CN110309687B - A two-dimensional code image correction method and correction device - Google Patents

Abstract

The invention discloses a correction method and a correction device of a two-dimensional code image, wherein the correction method comprises the steps of processing the two-dimensional code image to be corrected to obtain a boundary outline of the two-dimensional code image to be corrected; processing the boundary contour by adopting a first algorithm to obtain a first group of vertex positions, and processing the boundary contour by adopting a second algorithm to obtain a second group of vertex positions; blending the vertex positions of the same kind in the first group of vertex positions and the second group of vertex positions to obtain a target vertex position; and correcting the two-dimensional code image to be corrected based on the target vertex position to obtain a target two-dimensional code image. According to the method, the accuracy of determining the vertex position can be improved through two algorithms, and finally correction is performed according to the target vertex position, so that the deformation of the two-dimensional code image can be corrected, and the two-dimensional code can be conveniently identified.

Description

A two-dimensional code image correction method and correction device

technical field

The invention belongs to the technical field of two-dimensional code identification, and more particularly, relates to a two-dimensional code image correction method and correction device.

Background technique

A two-dimensional code is a black and white pattern distributed on a plane (two-dimensional direction) with a certain specific geometric pattern according to a certain rule to record data symbol information; in the code compilation, the "0" which constitutes the internal logic foundation of the computer is cleverly used. ”, “1” bit stream concept, using several geometric shapes corresponding to binary to represent text and numerical information, which can be automatically read by image input device or photoelectric scanning device to realize automatic information processing. Two-dimensional code has some common features of barcode technology: each code system has its specific character set; each character occupies a certain width; has a certain verification function, etc. At the same time, it also has the characteristics of automatic identification of information of different lines and processing of graphics rotation changes.

Commonly used two-dimensional code systems are: PDF417, Data Matrix, QR Code, Code 49, Code16K, Codeone, etc. It has the characteristics of large information capacity, wide coding range, strong fault tolerance, high decoding reliability, low cost, high confidentiality and good anti-counterfeiting performance. It plays an important role in information acquisition, website redirection, advertisement push, mobile e-commerce, anti-counterfeiting traceability, preferential promotion, membership management and mobile payment.

However, in the process of printing the QR code, due to equipment accuracy and other reasons, the edges of the QR code will be blurred and deformed to varying degrees. It is difficult to achieve positioning and recognition, resulting in too long recognition time or even inability to recognize, affecting the QR code. usage of.

In view of this, overcoming the defects of the prior art is an urgent problem to be solved in the technical field.

SUMMARY OF THE INVENTION

In view of the above defects or improvement requirements of the prior art, the present invention provides a two-dimensional code image correction method and correction device, the purpose of which is to obtain two sets of vertex positions of the two-dimensional code image through two different algorithms, and to The similar vertices in the two sets of vertex positions are reconciled to obtain the target vertex position, thereby determining the boundary point of the two-dimensional code image. The accuracy of vertex position determination can be improved through two algorithms, and finally the correction is performed according to the target vertex position, which can be corrected. The deformation of the two-dimensional code image to facilitate the identification of the two-dimensional code.

In order to achieve the above object, according to an aspect of the present invention, a method for calibrating a two-dimensional code image is provided, and the method for calibrating includes:

processing the two-dimensional code image to be corrected to obtain the boundary contour of the two-dimensional code image to be corrected;

The first algorithm is used to process the boundary contour to obtain a first set of vertex positions, and the second algorithm is used to process the boundary contour to obtain a second set of vertex positions;

Reconcile the vertex positions of the first group with the same vertex positions in the second set of vertex positions to obtain the target vertex position;

Correcting the to-be-corrected two-dimensional code image based on the target vertex position to obtain a target two-dimensional code image.

Preferably, after using the first algorithm to process the boundary contour to obtain the first group of vertex positions, and using the second algorithm to process the boundary contour to obtain the second group of vertex positions, the method further includes:

obtaining the similarity between the first set of vertex positions and the second set of vertex positions;

When the similarity between the first group of vertex positions and the second group of vertex positions is not less than a preset similarity threshold, perform a comparison between the first group of vertex positions and the same type of vertices in the second group of vertex positions The steps of reconciling the positions to get the target vertex position.

Preferably, the first set of vertex positions includes four vertex positions, the second set of vertex positions includes four vertex positions, and the distance between the first set of vertex positions and the second set of vertex positions is obtained. Similarities include:

respectively acquiring the first quadrilateral formed by the first group of vertex positions and the second quadrilateral formed by the second group of vertices;

respectively acquiring the area of the intersection area formed by the first quadrilateral and the second quadrilateral, and the area of the union area formed by the first quadrilateral and the second quadrilateral;

The intersection ratio of the area of the intersection area and the area of the merge area is calculated, and the similarity between the first set of vertex positions and the second set of vertex positions is marked by the intersection ratio.

Preferably, the first algorithm is an LSD line segment detection algorithm, the first set of vertex positions includes four vertex positions, and the use of the first algorithm to process the boundary contour to obtain the first set of vertex positions includes:

Using the LSD line segment detection algorithm to process the boundary contour to obtain a plurality of boundary lines;

Determine whether the number of the boundary lines is greater than four, and if the number of the boundary lines is greater than four, then mark the longest boundary line among the plurality of boundary lines as the target boundary line;

Traverse the lengths of the remaining boundary lines in turn, select the longest one as the boundary line to be verified, and calculate whether the distance between the midpoint of the boundary line to be verified and the midpoint of the target boundary line is less than a preset distance threshold;

If it is not less than, mark the boundary line to be verified as the target boundary line, if it is less than, remove the boundary line to be verified until four target boundary lines are screened;

Obtain the intersection of the four target boundary lines, and obtain the four vertex positions, thereby obtaining the first set of vertex positions.

Preferably, the second algorithm is Hough transform, the second set of vertex positions includes four vertex positions, and the second algorithm is used to process the boundary contour to obtain the second set of vertex positions including:

Hough transform is performed on the boundary contour to obtain boundary line segment clusters;

Perform slope and intercept binary clustering on the boundary line segment clusters to obtain upper boundary line segment clusters, lower boundary line segment clusters, left boundary line segment clusters and right boundary line segment clusters;

Perform straight line fitting on the upper boundary line segment cluster, the lower boundary line segment cluster, the left boundary line segment cluster and the right boundary line segment cluster respectively, and obtain four target boundary lines;

Obtain the intersection points of the four target boundary lines, and obtain the four vertex positions, thereby obtaining the second set of vertex positions.

Preferably, the number of the target vertices is four, and the two-dimensional code image to be corrected is corrected based on the position of the target vertex, and obtaining the target two-dimensional code image includes:

determining the position of the target vertex on the two-dimensional code image to be corrected to determine the boundary point of the two-dimensional code image to be corrected;

Perform perspective transformation on the to-be-corrected two-dimensional code image, correct the deformation of the to-be-corrected two-dimensional code image, and obtain a target two-dimensional code image.

Preferably, the calibration method further includes:

Use the third algorithm to perform edge detection on the target two-dimensional code image to obtain a first edge contour map, and use the fourth algorithm to perform edge detection on the target two-dimensional code image to obtain a second edge contour map;

Fusing the first edge contour map and the second edge contour map to obtain a target edge contour map;

The target two-dimensional code image is sharpened through the target edge contour map to enhance the edge of the target two-dimensional code image.

Preferably, the third algorithm is a Canny edge detection algorithm, and the fourth algorithm is a Sobel edge detection algorithm;

The first edge contour map and the second edge contour map are fused to obtain the target edge contour map including:

The first edge contour map and the second edge contour map are pixel superimposed according to a preset ratio to obtain a target edge contour.

Preferably, the processing of the to-be-corrected two-dimensional code image to obtain the boundary contour of the to-be-corrected two-dimensional code image includes:

Binarize the QR code image to be corrected;

Morphological corrosion is performed on the two-dimensional code image to be corrected that has undergone the binarization process to obtain the boundary contour of the two-dimensional code image to be corrected.

According to another aspect of the present invention, there is provided a correction device, comprising at least one processor; and, a memory communicatively connected to the at least one processor; wherein the memory stores information that can be used by the at least one processor Instructions to be executed, the instructions are programmed to perform the calibration method of the present invention.

In general, compared with the prior art, the above technical solutions conceived by the present invention have the following beneficial effects: the present invention provides a two-dimensional code image correction method and correction device, the correction method includes the two-dimensional code to be corrected. The image is processed to obtain the boundary contour of the two-dimensional code image to be corrected; the first algorithm is used to process the boundary contour to obtain the first group of vertex positions, and the second algorithm is used to process the boundary contour to obtain the second group of vertex positions; The first group of vertex positions and the similar vertex positions in the second group of vertex positions are reconciled to obtain the target vertex position; based on the target vertex position, the two-dimensional code image to be corrected is corrected to obtain the target two-dimensional code image. The invention obtains two sets of vertex positions of the two-dimensional code image through two different algorithms, and reconciles the same vertices in the two sets of vertex positions to obtain the target vertex position, thereby determining the boundary points of the two-dimensional code image. The algorithm can improve the accuracy of vertex position determination, and finally correct according to the target vertex position, which can correct the deformation of the two-dimensional code image, so as to facilitate the identification of the two-dimensional code.

Further, the edge contour image of the two-dimensional code image is obtained through multi-operator edge detection, so as to sharpen the two-dimensional code image and enhance the edge contour of the two-dimensional code image, which can eliminate the influence of the blurred edge of the two-dimensional code and obtain the background. A QR code image with clean, clear barcode and strong contrast can effectively improve the accuracy of subsequent identification.

Description of drawings

In order to describe the technical solutions of the embodiments of the present invention more clearly, the following briefly introduces the accompanying drawings that need to be used in the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention, and for those of ordinary skill in the art, other drawings can also be obtained from these drawings without creative effort.

1 is a schematic flowchart of a method for correcting a two-dimensional code image according to an embodiment of the present invention;

2 is a schematic diagram of a boundary outline of a two-dimensional code image to be corrected according to an embodiment of the present invention;

3 is a schematic diagram of converting a boundary line segment cluster into a boundary line provided by an embodiment of the present invention;

4 is a schematic structural diagram of an intersection area and a merge area between two quadrilaterals provided by an embodiment of the present invention;

FIG. 5 is a schematic structural diagram of the merged region of two quadrilaterals in FIG. 4 according to an embodiment of the present invention;

6 is a schematic structural diagram of an intersection area of two quadrilaterals in FIG. 4 according to an embodiment of the present invention;

7 is a schematic structural diagram of a target vertex position obtained by reconciling a first group of vertex positions and a second group of vertex positions according to an embodiment of the present invention;

8 is a schematic diagram of a corrected two-dimensional code image provided by an embodiment of the present invention;

FIG. 9 is a schematic flowchart of one of the implementation manners of step 102 in FIG. 1 provided by an embodiment of the present invention;

10 is a schematic diagram of the distribution of a boundary line segment cluster provided by an embodiment of the present invention;

11 is a schematic flowchart of another method for calibrating a two-dimensional code image provided by an embodiment of the present invention;

12 is a schematic diagram of a first edge contour diagram provided by an embodiment of the present invention;

13 is a schematic diagram of a second edge contour diagram provided by an embodiment of the present invention;

FIG. 14 is a schematic structural diagram of a calibration apparatus provided by an embodiment of the present invention.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, but not to limit the present invention.

In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

Example 1:

In practical application scenarios, due to the accuracy of the printing equipment or the angle of the camera, the obtained QR code image is often deformed, which affects subsequent QR code recognition. In order to solve the aforementioned problems, this embodiment provides a method for correcting a two-dimensional code image, which can correct a deformed two-dimensional code image, and convert the two-dimensional code image into a standard shape image, so as to facilitate the subsequent two-dimensional code image. identify.

Referring to FIG. 1 , the method for calibrating a two-dimensional code image in this embodiment includes the following steps:

In step 101, the two-dimensional code image to be corrected is processed to obtain the boundary contour of the two-dimensional code image to be corrected.

Among them, the two-dimensional code image to be corrected is mainly a two-dimensional code image with deformation or distortion. The deformation of the two-dimensional code may be caused by the printing equipment, causing the two-dimensional code image to cause deformation or distortion, and the deformation of the two-dimensional code may also be It is caused by a tilted shooting angle when a user takes a photo or scans a QR code with a smart device.

Among them, the code system of the QR code image to be corrected includes: PDF417, Data Matrix, QR Code, Code 49, Code16K or Code one, etc.

Before acquiring the boundary contour of the two-dimensional code image to be corrected, grayscale processing and binarization processing are performed on the two-dimensional code image to be corrected, and then the boundary contour is segmented by morphological corrosion processing.

As shown in FIG. 2 , the picture on the left is the image of the two-dimensional code to be corrected, and the picture on the right is the boundary outline of the image of the two-dimensional code to be corrected.

In step 102, a first algorithm is used to process the boundary contour to obtain a first set of vertex positions, and a second algorithm is used to process the boundary contour to obtain a second set of vertex positions.

In this embodiment, as shown in Figure 3, two different algorithms are used to process the boundary contour to obtain a boundary line segment cluster (the figure on the left in Figure 3), and the target boundary line is obtained according to the boundary line segment cluster (Figure 3 The figure on the right), according to the intersection between the target boundary lines, the first group of vertex positions and the second group of vertex positions are obtained respectively, wherein each group of vertex positions includes the position coordinates of multiple vertices, and the vertices can be understood as The boundary point of the two-dimensional code image, the shape and area size of the two-dimensional code image can be determined by the vertex.

In practical application scenarios, the two-dimensional code image is generally in the shape of a quadrilateral. In an optional embodiment, each group of vertex positions includes the position coordinates of the four vertices, wherein the intersection between the four boundary lines enclosing the quadrilateral is is the vertex (boundary point) of the QR code image.

In an optional solution, the first algorithm is LSD (Line Segment Detector, LSD for short) straight line detection, and the second algorithm is Hough transform. In other solutions, the specific types of the first algorithm and the second algorithm may also be other algorithms, which are not specifically limited here.

In step 103, a target vertex position is obtained by reconciling the vertex positions of the first set with the vertex positions of the same type in the second set of vertex positions.

Wherein, the vertex positions of the same type refer to two vertices that are closest to each other in the first set of vertex positions and the second set of vertex positions.

Wherein, the vertex position may be determined according to the coordinates in the preset coordinate system.

In an optional embodiment, in a preset coordinate system, the coordinates of the target vertex are obtained by performing harmonic averaging according to the horizontal and vertical coordinates of vertices of the same type.

In step 104, the to-be-corrected two-dimensional code image is corrected based on the target vertex position to obtain a target two-dimensional code image.

In this embodiment, the position of the target vertex is determined on the two-dimensional code image to be corrected to determine the boundary point of the two-dimensional code image to be corrected; then, the perspective of the two-dimensional code image to be corrected is performed Transform, correct the deformation of the two-dimensional code image to be corrected, and obtain the target two-dimensional code image.

The invention obtains two sets of vertex positions of the two-dimensional code image through two different algorithms, and reconciles the same vertices in the two sets of vertex positions to obtain the target vertex position, thereby determining the boundary points of the two-dimensional code image. The algorithm can improve the accuracy of vertex position determination, and finally correct according to the target vertex position, which can correct the deformation of the two-dimensional code image, so as to facilitate the identification of the two-dimensional code.

In practical application scenarios, although the specific methods of obtaining the vertex positions of the first algorithm and the second algorithm are different, for the same two-dimensional code image to be corrected, no matter which algorithm targets the same vertex, the obtained vertex positions will not be very different. Otherwise, it may be caused by errors in the identification and calibration of the vertex positions. In order to avoid errors in the identification of the vertex positions due to the problem of the algorithm itself and improve the accuracy, in a preferred embodiment, it is necessary to perform similarity matching between the positions of the first group of vertices and the positions of the second group of vertices. Perform the next steps.

This embodiment provides another method for calibrating a two-dimensional code image. After step 102, the method further includes acquiring the similarity between the first group of vertex positions and the second group of vertex positions. When the similarity between the vertex positions of the group and the vertex positions of the second group is not less than the preset similarity threshold, step 103 is performed again.

In an optional embodiment, the degree of similarity between the first set of vertex positions and the second set of vertex positions may be obtained by means of an intersection ratio.

Specifically, the first set of vertex positions includes four vertex positions, the second set of vertex positions includes four vertex positions, and the similarity between the first set of vertex positions and the second set of vertex positions is obtained The steps include: respectively acquiring a first quadrilateral formed by the first group of vertex positions and a second quadrilateral formed by the second group of vertices; acquiring the first quadrilateral and the second quadrilateral respectively the area of the intersection area formed by the polygons, and the area of the union area formed by the first quadrilateral and the second quadrilateral; calculate the intersection of the area of the intersection area and the area of the union area Union ratio, the similarity between the first set of vertex positions and the second set of vertex positions is marked by the intersection and union ratio.

The preset similarity threshold value may be an intersection ratio of 0.8, that is, when the intersection ratio of the area S1 of the _first quadrilateral and the area S2 of the _second quadrilateral is not less than 0.8 , the first set of vertex positions and the second set of vertex positions satisfy the similarity condition. Of course, the preset similarity threshold may also be other values, which are not specifically limited here.

The intersection ratio IOU is calculated according to the following formula: IOU=(S ₁ ∩ S ₂ )/(S ₁ ∪ S ₂ ).

In the actual application scenario, the calculation process of the intersection-union ratio IOU is as follows: (1) According to the vertex coordinates in the first group of vertex positions and the second group of vertex positions, calculate the first quadrilateral and the second quadrilateral respectively. boundary line, and determine the intersection point between the boundary line of the first quadrilateral and the boundary line of the second quadrilateral; (2) determine the inner and outer points of the first quadrilateral and the second quadrilateral respectively , where the interior point is defined as the boundary point (vertex) of one quadrilateral inside another quadrilateral, and the exterior point is defined as the boundary point (vertex) of one quadrilateral outside the other quadrilateral; (3) Connect the interior point and the intersection point in turn The intersection area is formed, and the outer points and the intersection points are connected in turn to form the merge area; (4) The intersection area and the merge area are divided into multiple triangles according to the boundary points (vertices), and the area sum of the triangles is calculated; (5) The area of the intersection area is higher than The area of the merged area is to be combined and compared.

Among them, the Heron formula can be used to obtain the sum of the areas of the triangles, and the Heron formula is as follows:

where a, b, and c are the side lengths of the triangle,

4 to 6, an example will be given. Here, it should be noted that, in order to clearly illustrate the implementation of this implementation, the two quadrilaterals shown in the accompanying drawings are deformed and the position difference is exaggerated. In practical application scenarios, the position difference between the quadrilaterals obtained by the two different algorithms is very small.

Assume that the two quadrilaterals for similarity determination are ABCD and A'B'C'D' respectively, and they intersect at points E and F. Among them, E and F are called intersection points, B and D' are inner points, and A', B', C', C, D, and A are called outer points.

Connect the intersection point and the outer point in turn to obtain the polygon A'B'C'FCDAE (as shown in Figure 5), which is called the union area. Take a certain point (it can be E) as a fixed point, and take two consecutive points clockwise (or counterclockwise) for the other points to form a triangle with E respectively, and solve the area of the triangle by Heron's formula to further solve the area area.

The intersection and interior points are connected in turn to obtain a quadrilateral EBFD' (as shown in Figure 6), which is called the intersection area. Take a certain point (it can be E) as a fixed point, and take two consecutive points clockwise (or counterclockwise) for the other points to form a triangle with E respectively, and then solve the area of the intersection area by solving the triangle area by Heron's formula.

In the process of program execution, it is necessary to execute "sequential connection" according to a strict algorithm to obtain the intersection area and merge area correctly. The following describes the algorithm implementation process in detail:

For the first polygon and the second polygon, the intersection point between the boundary lines of the two is the intersection point, the first polygon includes a plurality of vertices, and the first polygon is located outside the second polygon The outer point of the second polygon is the inner point of the first polygon; the second polygon includes a plurality of vertices, and the outer point of the second polygon is located outside the first polygon, Inside the first polygon is the interior point of the second polygon.

The determination process of the intersection area: take any point of the first polygon as the starting point, and record the first point sequence corresponding to the interior point of the first polygon and the intersection point according to the preset direction, wherein the preset direction is the order Clockwise or counterclockwise; take any point of the second polygon as the starting point, and record the second point sequence corresponding to the interior points and intersection points of the second polygon according to the preset direction, wherein the preset direction is clockwise Or counterclockwise. Preset comprehensive processing is performed on the first point sequence and the second point sequence to obtain the connection sequence of points on the intersection area, thereby obtaining the intersection area. Specifically, the intersection point sequence is determined in the first point sequence, the interior point sequence in the intersection point sequence is determined in the second point sequence, and the interior point sequence is inserted between the intersection point sequences in the first point sequence to obtain the intersection point sequence. The connection order of points on the area.

The process of judging the merged area: take any point of the first polygon as the starting point, and record the third point sequence corresponding to the outer point and the intersection point of the first polygon according to the preset direction, wherein the preset direction is the order Clockwise or counterclockwise; take any point of the second polygon as the starting point, and record the fourth point sequence corresponding to the outer point and the intersection point of the second polygon according to the preset direction, wherein the preset direction is clockwise Or counterclockwise. Preset comprehensive processing is performed on the third point sequence and the fourth point sequence, and the connection sequence of the points on the merged area is obtained, thereby obtaining the merged area. Specifically, the intersection point sequence is determined in the third point sequence, the outer point sequence in the intersection point sequence is determined in the fourth point sequence, and the outer point sequence is inserted between the intersection point sequences in the third point sequence to obtain the The connection order of points on the area.

In the following, in conjunction with FIG. 4 , how to obtain the intersection area and the merge area when the similarity is discriminated will be described in detail. Here, it should be noted that, in this embodiment, a quadrilateral is used as an example for illustration, and in an actual application scenario, the foregoing algorithm process is applicable to any polygon.

As shown in FIG. 4 , it is assumed that the two quadrilaterals for similarity determination are the first quadrilateral ABCD and the second quadrilateral A′B′C′D′ respectively, and the two intersect at points E and F. Among them, E and F are called intersection points, B is the interior point of the first quadrilateral, D' is the interior point of the second quadrilateral, and A', B', and C' are the exterior points of the second quadrilateral , C, D, and A are the outer points of the first quadrilateral.

Intersection area determination: take any point on the first quadrilateral ABCD as the starting point, record the sequence of interior points and intersection points clockwise, which can be EBF (first point sequence); take the second quadrilateral A'B' as the starting point Any point on C'D' is the starting point, and the sequence of interior points and intersection points on it is recorded clockwise, which can be FD'E (second sequence of points). By synthesizing the first point sequence and the second point sequence, the connection sequence of the points on the intersection area is obtained as EBFD'. Specifically, in the first point sequence, EF is the intersection point sequence, and in the second point sequence, D' is the interior point sequence between the intersection point sequences. D' is inserted between the intersection point sequences FE of the first point sequence to obtain EBFD'. Here, it should be noted that the intersection point sequence EF and the intersection point sequence FE are substantially the same intersection point sequence, and it is sufficient to ensure that the cycle can be performed in the manner of EFEF.

Combined area determination: take any point on the first quadrilateral ABCD as the starting point, record the point sequence of the outer points and intersection points clockwise, which can be AEFCD (third point sequence);

Taking any point on the second quadrilateral A'B'C'D' as the starting point, record the point sequence of the outer points and intersection points clockwise, which can be A'B'C'FE (the fourth point sequence). Combining the third point sequence and the fourth point sequence, the connection sequence of the points on the merged region is obtained as AEA'B'C'FCD. Specifically, in the third point sequence, EF is the intersection point sequence, and in the fourth point sequence, A'B'C' is the outer point sequence between the intersection point sequences, according to the intersection point sequence cycle (EFEF... ...), insert A'B'C' into the intersection point sequence EF of the first point sequence to obtain AEA'B'C'FCD. The above-mentioned two intersection points are used as an example. In a specific application scenario, the number of intersection points may be more than two. Take two consecutive intersection points in turn (note that it is a cyclic point sequence) and process it according to the above. , and will not be repeated here.

The method of this embodiment can be well implemented by a program, so that the computer can understand the meaning of sequential connection, so as to ensure that the intersection area and the merge area can be accurately determined, and the similarity determination can be completed.

After passing the similarity judgment, the harmonic mean of the two points of the same kind is used as the final boundary point, and the two points of the same kind are defined as the points located in the adjacent azimuths, that is, the two nearest points; if they do not pass, the correction fail. where the harmonic mean c of numbers a and b is defined as:

As shown in Figure 7, it is the position of the quadrilateral obtained by the two algorithms in the actual situation, A, B, C, D are the boundary points (vertices) of one of the quadrilaterals, A', B', C', D' is the boundary point (vertex) of another quadrilateral. Among them, A and A', B and B', C and C', D and D' are called similar points. According to the weighted average of the horizontal and vertical coordinates of similar points, the four boundary points of the new quadrilateral are obtained, that is, the graph In a, b, c, d, among them, a, b, c, d are the target vertices.

As shown in FIG. 8 , it is the two-dimensional code image after the two-dimensional code image to be corrected is corrected based on the target vertex position, that is, the target two-dimensional code image obtained in step 104 .

In step 101, processing the two-dimensional code image to be corrected, and obtaining the boundary contour of the two-dimensional code image to be corrected specifically includes: after performing grayscale processing on the two-dimensional code image to be corrected, binary value processing of the two-dimensional code image to be corrected processing. Then, perform multiple morphological erosion operations on the two-dimensional code image to be corrected to obtain the boundary contour of the two-dimensional code image to be corrected.

In an optional solution, the image of the two-dimensional code to be corrected is subjected to n times of morphological erosion operations in turn, and then the difference is made with the image after the n+1th morphological corrosion operation to obtain the boundary contour of the image of the two-dimensional code to be corrected, Among them, n can be 5 or other values. In this embodiment, through the morphological erosion operation, the area of the black area of the image is increased and the area of the white area of the image is reduced, so that the boundary contour of the two-dimensional code image to be corrected can be effectively extracted.

Among them, the specific method of morphological corrosion is: first set a square structural element S (can be 5 × 5), use the structural element S to erode the two-dimensional code image X, if the structural element is moved by x and included in the image foreground X ₀ , then it still belongs to the image foreground X ₀ , otherwise it belongs to the image background X ₁ , wherein the image foreground X ₀ refers to the white area and the image background X ₁ refers to the black area. The formula is as follows:

Among them, x is the vector distance that the structuring element moves, and the definition domain of x is the distance vector from all points on the foreground image X ₀ to the origin.

In an optional solution, the maximum inter-class variance method (Otsu) can be used for binarization, and the two-dimensional code image X is divided into an image foreground X ₀ and an image background X ₁ . Specifically, the size of the two-dimensional code image is M·N, the number of pixels whose gray value is less than the threshold value T in the two-dimensional code image is recorded as N0 (foreground pixels, forming the image foreground X ₀ ), and the gray value of the pixels is greater than the threshold value T The number of pixels is denoted as N ₁ (background pixels, forming the image background X ₁ ). Then there are: N ₀ +N ₁ =M·N.

Also note that the proportion of the image occupied by the foreground pixels is w ₀ , and the average gray level is u ₀ ; the proportion of the image occupied by the background pixels is w ₁ , and the average gray level is u ₁ . The average gray level of the image is u, and the inter-class variance is g, then:

w ₀ =N ₀ /(M·N) w ₁ =N ₁ /(M·N) w ₀ +w ₁ =1

u=w ₀ ·u ₀ +w ₁ ·u ₁ g=w ₀ ·(u ₀ -u) ² +w ₁ ·(u ₁ -u) ²

The equivalent formula is obtained: g=w ₀ ·w ₁ ·(u ₀ -u ₁ ) ² , and the threshold T that maximizes the variance between classes is obtained by traversing, which is the desired binarization threshold.

In this embodiment, the larger the inter-class variance between the background and the foreground, the larger the difference between the two parts constituting the image, and the smaller the misclassification probability.

Further, the two-dimensional code image to be corrected is binarized according to the obtained binarization threshold, and the two-dimensional code image to be corrected X is divided into an image foreground X ₀ and an image background X ₁ .

In an optional embodiment, the first algorithm is an LSD line segment detection algorithm, the first group of vertex positions includes four vertex positions, and the first algorithm is used to process the boundary contour to obtain four boundary lines , find the intersection of the four boundary lines, and obtain the coordinates of the four vertices, thereby obtaining the first set of vertex positions.

The following is a detailed description of the process of performing LSD line segment detection on the boundary contour, which mainly includes detecting short straight lines in the boundary contour, and then splicing the connected short straight lines to obtain a boundary line. The LSD line segment detection in this embodiment is improved based on the traditional detection method, and specifically includes the following steps:

其中，X*S为击中击不中变换，解释如下：设X^C是边界轮廓X的补集，结构元素S是由两个互不相交的部分S₁和S₂组成，即S＝S₁∪S₂且S₁∩S₂＝Φ(Φ代表空集)。则击中击不中变换定义为X*S＝(XΘS₁)∩(X^CΘS₂)，Θ为形态学腐蚀运算。(1) Perform multiple morphological refinements on the boundary contour. The specific method is as follows: set a square structural element S (can be 3×3), and use the structural element S to perform multiple refinement operations on the boundary contour X. The operation formula is:

Among them, X*S is the hit-miss transformation, which is explained as follows: Let X ^C be the complement of the boundary contour X, and the structuring element S is composed of two disjoint parts S ₁ and S ₂ , that is, S=S ₁ ∪S ₂ and S ₁ ∩S ₂ =Φ (Φ represents the empty set). Then the hit-miss transformation is defined as X*S=(XΘS ₁ )∩(X ^C ΘS ₂ ), and Θ is the morphological erosion operation.

(2) Line segment connection points in the fracture boundary contour. After edge thinning of the boundary contour, the number of neighbors in the 8 connected regions of edge pixels should be defined in the following cases: a pixel with two neighbors is the midpoint of a line segment; a pixel with one neighbor is an endpoint of a line segment ; Pixels with more than two neighbors are junctions connecting line segments. Break a line segment with a connection point at the connection point to obtain an independent boundary line segment.

(3) Set the Gaussian kernel scale s (which can be 0.8) to perform Gaussian downsampling on the boundary contour, and calculate the gradient value and gradient direction of each point. All points are pseudo-sorted according to the gradient value, and a state list is established. First, all the points are preset to the UNUSED state (indicating that they have not been processed), and then the state of the points whose gradient value is less than a specific value ρ is set to the USED state (indicating that they have been processed). , update the state list, where the specific value ρ is related to the angle tolerance, ρ=2/sin(t). In order to ensure that all gradient values are determined by the following step (4) to determine whether they are points on the boundary line segment, where the value less than ρ is directly set as USED because the gradient value is too small, it is impossible to be a boundary line segment, and no need to go through determination.

(4) Take the point with the largest gradient in the state list as the seed point, and set the state as USED. Starting from the seed point, search for points within the surrounding angle tolerance range [-t, t], change the status of the points that meet the angle tolerance range to USED, and generate a rectangle R containing all the points. Determine whether the inner point density of the rectangle R is greater than the threshold D, where the threshold D can be 0.2 or other data; if the point density is not greater than the threshold D, the rectangle R is truncated, and multiple rectangles are generated until the point density in the rectangle is greater than the threshold D. .

(5) Calculate the NFA (Number of False Alarms) of the rectangle R. If it is less than the threshold ε, add the rectangle R to the output list; otherwise, change the rectangle R until the NFA is less than the threshold ε, where the threshold ε can be 1 or other data. The NFA calculation formula is as follows:

NFA=N·PhO[k(r,I)≥k(r,i)]

Among them, N is the number of line segments in the current size m·n boundary contour image, which is (m·n) ^2.5 . k(r,I) is the pre-established contrast model, and the number of aligned points in the rectangle R in I. k(r,i) is the observed image (observe image), that is, the number of aligned pts in the rectangle R in the rectangle R, i.

(6) Divide all rectangles R into two categories according to the length threshold, one is long rectangle, if the aspect ratio is greater than the threshold γ (where the threshold γ can be 10), the corresponding rectangle will be added to the direct candidate rectangle; if If the aspect ratio is smaller than the threshold γ, it will be cut along the midpoint of the long side and added to the rectangular R set for re-judgment. The other type is short rectangles, which can be used as candidate rectangles, part of curves or short line segments, and the short line segments whose length is less than the threshold d are deleted, where the threshold d can be 10 or other values.

(7) The points in the short rectangle are fitted by the least squares method. If the average distance between the points in the short rectangle and the fitted straight line is less than the threshold μ (where the threshold μ can be 0.5), it is regarded as a candidate rectangle. Otherwise, delete the short rectangle.

(8) In the straight line merging stage, any two candidate rectangles I and J are merged, and the least squares fitting is performed, and the estimated variance is calculated. The formula for the estimated variance is as follows:

Among them, x _i is the point in the two candidate rectangles, L is the fitted line equation, and N is the total number of points in the two candidate rectangles.

If they belong to the same straight line, the estimated variance should be less than the variance threshold, otherwise, the points in the two candidate rectangles are rejected to be fitted to a straight line.

Among them, the rejection area (variance threshold) is as follows:

Among them, n _i and n _j are the number of points in the two candidate rectangles I and J respectively, μ can take a value of -2, and σ ₀ ² is 1.

(9) Perform least squares fitting on the points in the candidate rectangle to obtain the final line cluster. The least squares fitting target is the smallest residual error, and the residual error calculation formula is:

Among them, a and b are the slope and intercept of the line to be determined, and x _i and y _i are the point coordinates. The final solution is:

According to the foregoing steps (1) to (9), the boundary line segment clusters corresponding to the two-dimensional code image to be corrected can be obtained.

In this embodiment, step (1), step (2), step (6), step (7), step (8), step (9) are improvements to the LSD line segment detection algorithm, step (1) Eliminates the situation that the excessively thick boundary contour is identified as multiple straight lines during straight line detection, so that the boundary determination is blurred; step (2) break the irregular line segments to ensure that the results of subsequent straight line detection meet the requirements to a greater extent; step (6) It has the function of optimizing the rectangle, delete the short rectangle in it, and break the polyline and curve in it; step (7) delete the unnecessary curve in the rectangle to ensure the correctness of the straight line detection result; steps (8), (9) ), which greatly reduces the number of boundary lines, making it feasible to solve vertex positions.

In the actual application scenario, in theory, there are only four boundary lines in the boundary line segment cluster, but at the same time, there may be four more boundary lines. In this case, the boundary line segment cluster needs to be screened to obtain four boundary lines. , and then obtain the first set of vertex positions according to the four boundary lines.

In an optional embodiment, referring to FIG. 9 , the optimization scheme of step 102 is as follows:

In step 1021, the boundary contour is processed by using the LSD line segment detection algorithm to obtain a plurality of boundary lines.

Among them, a plurality of boundary lines form the aforementioned boundary line segment clusters.

In step 1022, it is determined whether the number of the boundary lines is greater than four, and if the number of the boundary lines is greater than four, the longest boundary line among the plurality of boundary lines is marked as the target boundary line.

In order to clearly illustrate this step, referring to FIG. 10 , the boundary line segment cluster includes 8 boundary lines, and each boundary line is sorted in descending order according to the length of the boundary line. The sequence number of the boundary line is 8.

In this embodiment, the boundary line 1 is first marked as the target boundary line.

In step 1023, traverse the lengths of the remaining boundary lines in turn, select the longest one as the boundary line to be verified, and calculate whether the distance between the midpoint of the boundary line to be verified and the midpoint of the target boundary line is less than a preset distance threshold .

In step 1024, if it is not less than, mark the boundary line to be verified as the target boundary line, and if it is less than, remove the boundary line to be verified until four target boundary lines are screened.

Among them, it is necessary to perform multiple traversal to filter out other target boundary lines.

The first traversal: traverse the length of the remaining boundary lines (boundary line 2 to boundary line 8), select boundary line 2 as the boundary line to be verified, and calculate the midpoint of the boundary line to be verified (boundary line 2) and the target boundary line (boundary line 2). Whether the distance between the midpoints of line 1) is less than a preset distance threshold. If not less than, mark the boundary line to be verified as the target boundary line, and if it is less than, remove the boundary line to be verified. The preset distance threshold is related to the size of the two-dimensional code. Assuming that the size of the two-dimensional code is m·n, the preset distance threshold may be 0.5·(m·n) ^0.5 .

Wherein, if the distance between the midpoint of the boundary line 2 and the midpoint of the target boundary line (boundary line 1 ) is not less than a preset distance threshold, the boundary line 2 is marked as the target boundary line.

The second traversal: traverse the length of the remaining boundary lines (boundary line 3 to boundary line 8), select boundary line 3 as the boundary line to be verified, and calculate the midpoint of the boundary line to be verified (boundary line 3) and the target boundary line (boundary line 3). Whether the distance between the midpoints of line 1 and boundary line 2) is less than a preset distance threshold. If not less than, mark the boundary line to be verified as the target boundary line, and if it is less than, remove the boundary line to be verified.

Wherein, if the distance between the midpoint of the boundary line 3 and the midpoint of the target boundary line (boundary line 2) is less than the preset distance threshold, the boundary line 3 is removed, wherein the midpoint of the boundary line to be verified and all the targets When the distance between the midpoints of the boundary lines is not less than the preset distance threshold, the corresponding boundary line to be verified is marked as the target boundary line, otherwise, the corresponding boundary line to be verified is eliminated.

The third traversal: traverse the length of the remaining boundary lines (boundary line 4 to boundary line 8), select boundary line 4 as the boundary line to be verified, and calculate the midpoint of the boundary line to be verified (boundary line 4) and the target boundary line (boundary line 4). Whether the distance between the midpoints of line 1 and boundary line 2) is less than a preset distance threshold. If not less than, mark the boundary line to be verified as the target boundary line, and if it is less than, remove the boundary line to be verified.

Wherein, if the distance between the midpoint of the boundary line 4 and the midpoint of the target boundary line (boundary line 1 and boundary line 2) is not less than the preset distance threshold, the boundary line 4 is marked as the target boundary line.

And so on, until four target boundary lines are filtered out.

In step 1025, the intersections of the four target boundary lines are obtained, and the four vertex positions are obtained, thereby obtaining the first group of vertex positions.

In this embodiment, screening is performed according to the distance between the midpoints of the boundary lines, which can effectively eliminate the boundary lines that are relatively close to each other, and effectively obtain the boundary line closest to the boundary contour of the two-dimensional code, which improves the detection accuracy.

In a specific application scenario, the second algorithm is Hough transform, the second set of vertex positions includes four vertex positions, the second algorithm is used to process the boundary contour, and the obtained second set of vertex positions includes: : perform Hough transform on the boundary contour to obtain the boundary line segment cluster; perform slope and intercept binary clustering on the boundary line segment cluster to obtain the upper boundary line segment cluster, the lower boundary line segment cluster, the left boundary line segment cluster and the right boundary line segment cluster; Perform straight line fitting on the upper boundary line segment cluster, the lower boundary line segment cluster, the left boundary line segment cluster and the right boundary line segment cluster, respectively, to obtain four target boundary lines; Group vertex positions.

In this embodiment, the RANSAC algorithm may be used to process the upper boundary line segment cluster, the lower boundary line segment cluster, the left boundary line segment cluster and the right boundary line segment cluster to obtain four boundary lines.

First, briefly explain the idea of Hough transform: read a binary image to be processed; obtain the source pixel data of the binary image space; quantify the polar coordinate system parameter space into a finite number of value intervals to equally divide or accumulate the lattice, namely ρ and theta, where ρ and theta are parameters in the parameter space (polar coordinate system), representing length and angle respectively.

Through the Hough transform algorithm, each pixel coordinate point P(x, y) is converted to the curve point of (ρ, theta), and accumulated to the corresponding grid data point; find the maximum Hough value, and set the threshold (wherein , if the size of the two-dimensional code is m·n, the threshold value can be between 0.1·(m·n) ^0.5 ~ 0.2·(m·n) ^0.5 ), and inversely transform to the image space to obtain four boundary lines.

In this embodiment, Hough transform is performed on the boundary contour to obtain a boundary line segment cluster, and then slope and intercept binary clustering is performed on the boundary line segment cluster to obtain an upper boundary line segment cluster, a lower boundary line segment cluster, and a left boundary line segment cluster. and the right boundary line segment cluster. Specifically, the upper boundary line segment cluster and the lower boundary line segment cluster are separated from the left boundary line segment cluster and the right boundary line segment cluster according to the slope, and then the upper boundary line segment cluster and the lower boundary line segment cluster are separated according to the intercept, and the left boundary line segment is separated. Clusters are distinguished from right boundary line segment clusters.

Before clustering, K-means outlier detection is performed first. The specific steps are as follows:

(1) The initial abnormal data set M is the whole point set, wherein the abnormal data set M contains all the point sets formed by the slope and intercept;

(2) arbitrarily select k objects (where k is 2) as the initial cluster center, and assign each object to the most similar cluster;

(3) Calculate the average value of the objects in the cluster, and calculate the average radius of the cluster;

(4) Obtain the candidate abnormal object set N, wherein the candidate abnormal object set N refers to the objects whose distance from the central object is greater than the average radius, and the clusters with only one object;

(5) Find the intersection M=M∩N with the previous abnormal data set as a new abnormal data set;

(6) Repeat the above steps until the abnormal data set converges or the maximum number of iterations is reached, and a sample is obtained.

Then, K-medoids clustering algorithm is selected for clustering to avoid the influence of outliers on the boundary line determination algorithm again. The K-mediods clustering algorithm is:

(1) Select the maximum and minimum values of the sample as the initial mediods (center point set);

(2) According to the principle of being close to the initial mediods, assign the remaining n-2 points to the class represented by the current best mediods;

(3) If the number of a certain class is 1, delete this class, and select the maximum and minimum values from another class as mediods;

(4) For all other points in the class except the mediods point, calculate the value of the criterion function in order when it is a new mediods, traverse all possibilities, and select the point corresponding to the minimum criterion function as the new mediods; among them, The criterion function is: the sum of the distances from all other points in the current mediods to the center point is the smallest;

(5) Repeat the process of (2) to (4) until all the mediods no longer change or the set maximum number of iterations has been reached.

In this embodiment, for the original K-mediods clustering algorithm, in step (1), different from the arbitrary values in the original K-mediods clustering algorithm, this embodiment uses the maximum value and the minimum value as the initial value mediods can speed up the clustering efficiency; different from the original K-mediods clustering algorithm, step (3) is added to prevent the influence of outliers.

After processing the slope and intercept through the aforementioned K-means outlier detection and K-medoids clustering algorithm, the upper boundary line segment cluster, the lower boundary line segment cluster, the left boundary line segment cluster and the right boundary line segment cluster are obtained. Then, using RANSAC The algorithm processes each boundary line segment cluster separately, and obtains four boundary lines.

Specifically, the RANSAC algorithm is used to achieve the goal by repeatedly selecting a set of random subsets in the data. The selected subsets are assumed to be intra-office points, and the following methods are used to verify:

(1) K-means++ clustering is performed on the original data set, which is divided into two types of data points;

(2) Select a point in each of the two categories to estimate the straight line model, which is definite and unique;

(3) Use the model obtained in (2) to test all other data, if the distance from a point to the line model is less than the threshold L, where the threshold L can be the distance of 10 (or other numbers) pixels, then Consider it also an in-office point;

(4) If the number of intra-office points of the model exceeds the threshold N, where the threshold N can be 80%, the model is considered reasonable, and then the final straight line is fitted by the least squares method of all intra-office points;

(5) If the number of intra-office points is less than the threshold N but greater than the number of intra-office points of the previous model, the model is retained, otherwise it is discarded;

(6) Repeat (2) to (5) until the maximum number of iterations is reached.

In this embodiment, the original RANSAC algorithm is improved, which is different from the original RANSAC algorithm, and the improvement of step (1) and step (4) can improve the efficiency of the algorithm.

In an optional embodiment, the K-means++ clustering algorithm process is as follows:

(1) randomly select a sample point from the original data set as the initial cluster center c ₁ ;

(2) For each sample point _xi in the original data set, calculate the shortest distance between it and the existing cluster center, which is represented by D(x);

(3) Select a new sample as the new cluster center. The principle of selection is: the point with larger D(x) is more likely to be selected as the cluster center; D(x) is reflected to the selected point In terms of probability, the method is: first add the distance between each point in the data set and the nearest cluster center point, and the sum is represented by sum(D(x)); in 0-sum(D(x)) Take a random value Random, then Random=Random-D(x), until Random≤0, the point at this time is the next cluster center;

(4) Repeat (2) to (3) until k cluster centers are selected;

(5) For each sample x _i in the data set, calculate its Euclidean distance to k cluster centers, and divide it into the cluster corresponding to the cluster center with the smallest distance;

(6) For each cluster, update the cluster center, and the update function is represented by F(x); by C _i to represent the set of data points in the cluster, and |C _i | to represent the number of data points in the cluster, then the update function F (x) is:

(7) Repeat (5) to (6) until the cluster center does not change, and the error square sum criterion function converges E(x); the goal is to minimize the mean square error E, and the criterion function E(x) is:

The initial dataset is expressed as: X={x ₁ , x ₂ ,...,x _n }, and the number of clusters is k.

In a practical application scenario, the intersection point can be taken as follows, so as to obtain the first set of vertex positions and the second set of vertex positions. In this embodiment, the four target boundary lines are processed according to the following formula to obtain the intersection point. For example, straight line L1: y=ax+b, straight line L2: y=cx+d, and the coordinates of the intersection point are:

In an optional solution, in step 104, correcting the two-dimensional code image to be corrected based on the target vertex position, and obtaining the target two-dimensional code image specifically includes: determining on the two-dimensional code image to be corrected The target vertex position is used to determine the boundary point of the two-dimensional code image to be corrected; perspective transformation is performed on the two-dimensional code image to be corrected, and the deformation of the two-dimensional code image to be corrected is corrected to obtain the target two-dimensional code image.

Among them, perspective transformation is a mapping transformation that converts the image from the original two-dimensional space to the three-dimensional space, and then from the three-dimensional space to another new two-dimensional space. Among them, the formula of the original two-dimensional space to three-dimensional space mapping is as follows:

The mapping formula from the three-dimensional space to the new two-dimensional space is:

x=x'/w'y=y'/w'

Here, the original two-dimensional space corresponds to the original image, and the new two-dimensional space corresponds to the new image. Among them, (u, v) are the points on the original image, corresponding to the coordinates (x, y) of the transformed new image points. Rewrite the previous transformation formula to get:

According to the aforementioned perspective transformation, the deformation of the to-be-corrected two-dimensional code image is corrected to obtain the target two-dimensional code image.

Different from the prior art, the present invention obtains two sets of vertex positions of the two-dimensional code image through two different algorithms, and reconciles the same vertices in the two sets of vertex positions to obtain the target vertex position, thereby determining the position of the two-dimensional code image. For boundary points, two algorithms can improve the accuracy of vertex position determination, and finally correct according to the target vertex position, which can correct the deformation of the two-dimensional code image, so as to facilitate the identification of the two-dimensional code.

Example 2:

The foregoing Embodiment 1 mainly describes the process of correcting a deformed two-dimensional code image. In practical application scenarios, the edges of the two-dimensional code image may be blurred, resulting in the inability to quickly and accurately identify the two-dimensional code image. To solve the aforementioned problems, this embodiment is further improved based on Embodiment 1. The edge contour image of the two-dimensional code image is obtained through multi-operator edge detection, so as to sharpen the two-dimensional code image and enhance the edge of the two-dimensional code image. The contour can eliminate the influence of the blurred edge of the two-dimensional code, and obtain a two-dimensional code image with a clean background, clear bar code and strong contrast, which can effectively improve the accuracy of subsequent identification.

Referring to FIG. 11 , this embodiment provides another method for calibrating a two-dimensional code image, and the method further includes the following steps:

Here, it should be noted that when the two-dimensional code image has problems of deformation and edge blurring, after correcting the image according to steps 101 to 104 of Embodiment 1, the target two-dimensional code image is obtained, and then according to the The method sharpens the edge of the target two-dimensional code image to eliminate the problem of edge blurring. When there is no deformation in the two-dimensional code image, but the edge of the two-dimensional code image is blurred, the method of this embodiment can be directly used to sharpen the edge of the target two-dimensional code image to eliminate the problem of blurred edges. That is, the correction method in this embodiment is suitable for different application scenarios.

In step 105, a third algorithm is used to perform edge detection on the target two-dimensional code image to obtain a first edge contour map, and a fourth algorithm is used to perform edge detection on the target two-dimensional code image to obtain a second edge contour map.

Wherein, the third algorithm is the Canny edge detection algorithm, and the fourth algorithm is the Sobel edge detection algorithm. The third algorithm and the fourth algorithm may also be other algorithms.

In this embodiment, two algorithms are used to perform multi-operator fusion. In an actual application scenario, when multi-operator fusion is performed, the number and type of algorithms may be more than two or more, which are not detailed here. Restriction, use different algorithms for edge detection in the aforementioned manner, and then fuse the edge contour maps corresponding to each algorithm according to the following steps to obtain the target edge contour map.

In practical application scenarios, before edge detection, the target QR code image is subjected to median filtering and anisotropy-based detail preservation algorithm for smoothing and denoising, and then multi-operator edge detection is performed.

In step 106, the first edge contour map and the second edge contour map are fused to obtain a target edge contour map.

In this embodiment, pixels of the first edge contour map and the second edge contour map are superimposed according to a preset ratio to obtain a target edge contour.

The preset ratio is 1:1 or other ratios, which are not specifically limited here.

In step 107, the target two-dimensional code image is sharpened by the target edge contour map to enhance the edge of the target two-dimensional code image.

In an optional solution, the principle of median filtering is to replace the value of a point in the digital image with the median value of each point in an area around the point. Take a 3×3 window as an example:

g(x,y)=median(f(x-1,y-1),f(x,y-1),f(x+1,y-1),f(x-1,y),

f(x,y),f(x+1,y),f(x-1,y+1),f(x,y+1),f(x+1,y+1))

Among them, f(x, y) is the pixel value of point (x, y), g(x, y) is the pixel value after median filtering, and median is the median function.

In an optional solution, the detail preservation algorithm based on anisotropy is specifically: In the traditional PM anisotropic diffusion model, it can be represented discretely by using the four nearest neighbors and the Laplacian operator, where ▽ I _i ^t (x, y) represents the gradient in four directions, c _i ^t (x, y) represents the diffusion coefficient associated with it, I _t (x, y) is the pixel value before processing, I _t+1 ( x, y) is the processed pixel value:

The function g(▽I) must be a monotonically decreasing function, satisfying g(0)=1, g(∞)=0, which can be expressed as (where K is a constant):

g(▽I)=1/[1+(|▽I|/K) ² ]

In practical application scenarios, if K is too large, the diffusion process will transition smoothly and cause image blurring; if K is too small, the diffusion process will stop smoothing in the early iteration process and produce a restored image similar to the original image. Therefore, consider an anisotropic diffusion model with gradient and variance as two local features. Taking a 3×3 area as an example, its grayscale variance:

Since the gray level has a larger variation range than the gradient, the variance needs to be normalized to make the two compatible:

Then the diffusion coefficient function is corrected as:

In an optional solution, the essence of performing multi-operator fusion edge detection on the image is: merging and superposing the first edge contour map obtained by the Canny edge detection algorithm and the second edge contour map obtained by the Sobel edge detection algorithm, Obtain the target edge contour map, in which the superposition ratio can be 0.5 and other values can be used, which can be set according to the actual situation.

The processing process of the Canny edge detection algorithm is described in detail below.

First, a brief description of the traditional steps of the Canny edge detection algorithm:

(1) Gaussian filtering to smooth the image

The Canny edge detection algorithm uses the derivative of a two-dimensional Gaussian function to process the original image, where (x,y) are the image point coordinates. Among them, the two-dimensional Gaussian function is as follows:

The gradient vector is:

(2) Calculate the magnitude and direction of the gradient

The partial derivatives along the x and y directions are _Px (i,j) and _Py (i,j) from the image I(x,y), respectively. Convert Cartesian coordinates to polar coordinates, and convert pixel P _x (i, j) and P _y (i, j) to gradient magnitude M(i, j) and gradient direction θ(i, j). M(i,j) represents the edge strength of any point, and θ(i,j) represents the normal vector of any point.

where _Px and _Py are defined as follows:

(3) Non-maximum suppression of gradient amplitude

Taking the 3×3 neighborhood points as the research object, it is found that the gradient directions of the neighborhood points are in the direction of the 0°, 45°, 90°, and 135° gradients of the sector. If the value is greater than the adjacent point, the point may be an edge point and can be reserved again for judgment. Otherwise, the point is not an edge point and the value is set to 0.

(4) Use dual thresholds to detect and connect edges

The value of point (i, j) is compared with the threshold, and if the gradient magnitude of the point is greater than the high threshold Th, it can be determined as an edge point. If Tl≤M(i,j)≤Th, wait for further judgment, and if the point gradient magnitude is less than the low threshold T1, it can be determined that it is not an edge point.

In further judgment, a high threshold is used to obtain an edge image first, but since the edge image adopts a high threshold, there are many cases of discontinuity. At this time, the 8 neighborhoods of all edge line endpoints in the high-threshold edge image are searched for edges that can be connected to the contour, and the connection is made when Tl≤M(i,j) is satisfied.

Since the traditional Canny edge detection algorithm is prone to the problems of detecting false edges and lost details, in a preferred embodiment, the traditional Canny edge detection algorithm is improved as follows:

The improved Canny edge detection algorithm uses multiple templates to convolve the original image f(x,y), where the original image refers to the aforementioned two-dimensional code image after smoothing and denoising. When the template affects the point, it is calculated according to the size of the gradient, the weighted value is constantly changing, and the Gaussian function G(x, y, σ) convolves f(x, y) to obtain a smooth image.

When the pixel value changes significantly, the filter weight can be set to 0 to avoid the smoothing of the pixel change and improve the computational efficiency. The characteristic of the improved adaptive filtering algorithm is that the weight coefficient is automatically adjusted when the weight coefficient satisfies the region without significant change.

Among them, k is the number of iterations, I ^(k) (x, y) is the image pixel value, the derivative of I (k ⁾ (x, y) is I′ ^(k) (x, y), Φ(I′ ^{(k )} (x, y)) is a monotonically decreasing function, the maximum value is Φ(0)=1, when I′ ^(k) (x, y) keeps increasing, Φ(I′ ^(k) (x, y)) reduced to 0. I' ^(k) (x, y) can detect whether there is a sudden change in the grayscale change, and w ^(k) (x, y) can be described as:

where h is a constant, and I′ ^(k) (x,y) is defined as:

Weight factor:

in,

In practical application scenarios, the inventor found that after 5 iterations, the edge sharpening effect was the best.

In this embodiment, the improved Canny edge detection algorithm is used to obtain the first edge contour map.

The processing process of the Sobel edge detection algorithm is described in detail below.

First, the traditional Sobel edge detection algorithm is as follows:

The gradient of the image f(x,y) at position (i,j) is defined as:

The magnitude of the gradient vector is:

In the classic Sobel operator, the horizontal and vertical direction template and image f(x,y) are convolved to approximate the gradient value, which can be solved by the following formula:

G _x =f(i+1,j-1)+2f(i+1,j)+f(i+1,j+1)-f(i-1,j-1)-2f(i-1 ,j)-f(i-1,j+1)

G _y =f(i-1,j+1)+2f(i,j+1)+f(i+1,j+1)-f(i-1,j-1)-2f(i,j -1)-f(i-1,j-1)

The processing process of the traditional Sobel edge detection algorithm is as follows: ① The horizontal and vertical bidirectional templates move from left to right, from top to bottom, and from one pixel to another along the image respectively, and the center point of the template corresponds to the corresponding pixel in the image ; ② Convert the weight of each position in the template to its corresponding image pixel value, and calculate the gradient value accordingly; ③ Replace the image pixel value corresponding to the center point of the template with the value of the gradient vector, expressed as px; ④ Set a suitable threshold T (can be 150), if px≥T, it is determined as the pixel point at the edge of the image.

Because the traditional Sobel operator is insensitive to edges in other directions, is not ideal for complex texture and oblique image contour extraction, and has low anti-noise ability, in a preferred embodiment, the traditional Sobel operator is improved as follows:

(1) Alternately smooth the image using the opening and closing operations in morphology;

(2) Use the extended Sobel operator for edge detection;

(3) Use the edge binary Otsu algorithm to select an appropriate threshold for binarization.

Among them, the morphological operation is defined as follows, where F(x,y) is the grayscale image, (x,y) is the image point coordinates, S(m,n) is the morphological structure element (the size can be 3×3), ( m,n) are the structuring element point coordinates:

Expansion:

Corrosion: FΘS=min{F(x+m,y+n)}

open:

closure:

Among them, compared with the traditional Sobel edge detection algorithm, the improved Sobel edge detection algorithm adds six directions, which are 45°, 135°, 180°, 225°, 270°, and 315°, which can obtain more accurate edge images.

In this embodiment, the second edge contour map is obtained through the Sobel edge detection algorithm.

In this implementation, the first edge contour map and the second edge contour map are fused to obtain a target edge contour map; the target two-dimensional code image is sharpened through the target edge contour map to Enhance the edge of the target QR code image.

Referring to FIG. 12 and FIG. 13 , the image on the right in FIG. 12 is the first edge contour image obtained by the Canny operator. It can be seen that the first edge contour map contains very detailed contours, but it is easy to misjudge noise as a boundary. The figure on the right in FIG. 13 is the second edge contour figure obtained by the Sobel operator. It can be seen that the effect of noise processing is better, but the edge location is not very accurate. In this embodiment, the first edge contour map and the second edge contour map are fused to obtain the target edge contour map, which can not only ensure detailed contours, accurate edge positioning, but also have better noise processing capabilities , which can effectively sharpen the QR code image.

In practical application scenarios, the sharpened target QR code image is further subjected to adaptive threshold binarization, and then the processed QR code image is sent to the recognition component for QR code recognition.

In an optional solution, the adaptive threshold binarization algorithm is introduced as follows:

The basic idea of the algorithm is to traverse the image, calculate a moving average, and set the pixel to black if it is significantly below this average, otherwise set to white. Suppose _Pn is the pixel at point n in the image, and fs(n) is the sum of the last _s pixels at point n:

Finally, the image T(n) is 1 (black) or 0 (white) depending on whether it is darker than t percent of the average of its first s pixels. Among them, s takes 1/8 of the width of the image and t takes 15 to achieve better results.

The calculation of the average can also take two sides symmetrical about the point n, or even replace the one-dimensional smoothed value with a two-dimensional smoothed value. Suppose P _i0,j0 is the pixel located at point (i ₀ ,j ₀ ) in the image, f _s (i ₀ ,j ₀ ) is the center of point (i ₀ ,j ₀ ), around (s+1)· Sum of (s+1) pixel values:

To calculate the sum of the pixel values in the specified rectangular area of the image, the image integral map can be used. Assuming that ii(x,y) represents the sum of the pixels in the upper left corner of the point, it is expressed as follows:

Then the sum of pixel values in the rectangular area represented by point (x ₀ , y ₀ ) and point (x, y) is expressed as:

S=ii(x,y)+ii(x ₀ ,y ₀ )-ii(x,y ₀ )-ii(x ₀ ,y)

In order to make the image pixel transition relatively smooth, traversing the image adopts a Z-shaped traversal. In order to make the threshold calculation take into account the information in the vertical direction, the average effect of the previous line is retained during traversal, and then the average of the current line and the average of the previous line is averaged as a new average.

Different from the prior art, the edge contour image of the two-dimensional code image is obtained through multi-operator edge detection, so as to sharpen the two-dimensional code image, enhance the edge contour of the two-dimensional code image, and eliminate the influence of the blurred edge of the two-dimensional code. , to obtain a QR code image with clean background, clear barcode and strong contrast, which can effectively improve the accuracy of subsequent identification.

Example 3:

In practical application scenarios, there may be errors in the identification of the vertex positions due to the problem of the algorithm itself. That is, in the above-mentioned Embodiment 1, the similarity between the first group of vertex positions and the second group of vertex positions is less than the preset threshold, in order to avoid If the correction is wrong or the correction time is too long, in this case, the vertex position obtained by one of the algorithms can be used as the target vertex position for image correction.

In practical application scenarios, although the detection speed of the LSD line segment detection algorithm is fast, when the image of the QR code to be corrected is blurred, the problem of low recognition rate is prone to occur. With reference to Embodiment 1, when the similarity between the first group of vertex positions and the second group of vertex positions is less than a preset similarity threshold, the total number of pixels per unit area of the two-dimensional code image to be corrected is obtained, wherein the unit area The total number of pixels in the QR code image can reflect the clarity of the QR code image to be corrected. When the total number of pixels per unit area of the QR code image to be corrected is greater than the preset number threshold, that is, the QR code image to be corrected is clearer. , the first group of vertex positions (obtained by the LSD line segment detection algorithm) can be used as the target vertex positions to determine the boundary points of the two-dimensional code image to be corrected, thereby avoiding correction errors or excessive correction time.

When the total number of pixels in the unit area of the two-dimensional code image to be corrected is not greater than the preset number threshold, that is, when the two-dimensional code image to be corrected is relatively blurred, the second group of vertex positions can be used as the target vertex positions (through the Huo obtained from the QR code image), and determine the boundary points of the two-dimensional code image to be corrected, so as to avoid correction errors or excessive correction time.

In an optional solution, the total number of pixels per unit area can be determined by the memory space occupied by the image per unit area, that is, acquiring the total number of pixels per unit area of the two-dimensional code image to be corrected specifically includes: pre-establishing the total number of pixels per unit area image. The mapping relationship between the occupied memory space and the total number of pixels per unit area, obtain the memory space occupied by the unit area image of the two-dimensional code image to be corrected, and calculate the per-unit area of the two-dimensional code image to be corrected based on the aforementioned mapping relationship the total number of pixels.

In this embodiment, in order to avoid the problem of correction errors caused by the algorithm, a variety of optional execution branches are provided. Preferably, the position of the target vertex is determined by the fusion of the two algorithms to improve the accuracy; When there is a large difference between the vertex positions determined by the algorithm, according to the total number of pixels in the unit area of the QR code image to be corrected, the LSD line segment detection algorithm or Hough transform is selectively selected to determine the vertex positions to avoid correction errors or notifications. The user captures the image again to improve the user experience of the user.

For other processes of the two-dimensional code image correction method, please refer to Embodiment 1 and Embodiment 2, which will not be repeated here.

Example 4:

Please refer to FIG. 14. FIG. 14 is a schematic structural diagram of a calibration apparatus provided by an embodiment of the present invention. The calibration apparatus of this embodiment includes one or more processors 41 and a memory 42 . Among them, one processor 41 is taken as an example in FIG. 14 .

The processor 41 and the memory 42 may be connected by a bus or in other ways, and the connection by a bus is taken as an example in FIG. 14 .

The memory 42 is used as a non-volatile computer-readable storage medium based on the two-dimensional code image correction method, and can be used to store non-volatile software programs, non-volatile computer-executable programs and modules, such as Embodiment 1 and /or the two-dimensional code image correction method and corresponding program instructions in Embodiment 2. The processor 41 executes various functional applications and data processing of the correction method for the two-dimensional code image by running the non-volatile software programs, instructions and modules stored in the memory 42 to implement Embodiment 1 and/or Embodiment 2 The function of the correction method of the QR code image.

The memory 42 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, memory 42 may optionally include memory located remotely from processor 41, which may be connected to processor 41 via a network. Examples of such networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

Regarding the correction method of the two-dimensional code image, please describe the relevant texts of Embodiment 1 and Embodiment 2 here, and will not repeat them.

It is worth noting that the information exchange, execution process and other contents between the modules and units in the above-mentioned device and the system are based on the same concept as the processing method embodiments of the present invention. For details, please refer to the descriptions in the method embodiments of the present invention. , and will not be repeated here.

Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the embodiments can be completed by instructing relevant hardware through a program, and the program can be stored in a computer-readable storage medium, and the storage medium can include: Read Only Memory (abbreviated as ROM), random access memory (Random Access Memory, abbreviated as RAM), magnetic disk or optical disk, etc.

Those skilled in the art can easily understand that the above are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present invention, etc., All should be included within the protection scope of the present invention.

Claims (8)

1. a correction method of two-dimensional code image, is characterized in that, described correction method comprises: processing the two-dimensional code image to be corrected to obtain the boundary contour of the two-dimensional code image to be corrected; The first algorithm is used to process the boundary contour to obtain a first set of vertex positions, and the second algorithm is used to process the boundary contour to obtain a second set of vertex positions; Reconcile the vertex positions of the first group with the same vertex positions in the second set of vertex positions to obtain the target vertex position; Correcting the to-be-corrected two-dimensional code image based on the target vertex position to obtain a target two-dimensional code image; After the first algorithm is used to process the boundary contour to obtain the first group of vertex positions, and the second algorithm is used to process the boundary contour to obtain the second group of vertex positions, the method further includes: obtaining the similarity between the first set of vertex positions and the second set of vertex positions; When the similarity between the first group of vertex positions and the second group of vertex positions is not less than a preset similarity threshold, perform a comparison between the first group of vertex positions and the same type of vertices in the second group of vertex positions The steps of reconciling the positions to obtain the target vertex position; The first set of vertex positions includes four vertex positions, the second set of vertex positions includes four vertex positions, and obtaining the similarity between the first set of vertex positions and the second set of vertex positions includes: : respectively acquiring the first quadrilateral formed by the first group of vertex positions and the second quadrilateral formed by the second group of vertices; respectively acquiring the area of the intersection area formed by the first quadrilateral and the second quadrilateral, and the area of the union area formed by the first quadrilateral and the second quadrilateral; The intersection ratio of the area of the intersection area and the area of the merge area is calculated, and the similarity between the first set of vertex positions and the second set of vertex positions is marked by the intersection ratio. 2. The correction method according to claim 1, wherein the first algorithm is an LSD line segment detection algorithm, the first group of vertex positions comprises four vertex positions, and the first algorithm is used to detect the boundary The contour is processed to obtain the first set of vertex positions including: Using the LSD line segment detection algorithm to process the boundary contour to obtain a plurality of boundary lines; Determine whether the number of the boundary lines is greater than four, and if the number of the boundary lines is greater than four, then mark the longest boundary line among the plurality of boundary lines as the target boundary line; Traverse the lengths of the remaining boundary lines in turn, select the longest one as the boundary line to be verified, and calculate whether the distance between the midpoint of the boundary line to be verified and the midpoint of the target boundary line is less than a preset distance threshold; If it is not less than, mark the boundary line to be verified as the target boundary line, if it is less than, remove the boundary line to be verified until four target boundary lines are screened; Obtain the intersection of the four target boundary lines, and obtain the four vertex positions, thereby obtaining the first set of vertex positions. 3 . The correction method according to claim 1 , wherein the second algorithm is Hough transform, the second group of vertex positions includes four vertex positions, and the second algorithm is used to analyze the boundary contour. 4 . Perform processing to obtain the second set of vertex positions including: Hough transform is performed on the boundary contour to obtain boundary line segment clusters; Perform slope and intercept binary clustering on the boundary line segment clusters to obtain upper boundary line segment clusters, lower boundary line segment clusters, left boundary line segment clusters and right boundary line segment clusters; Perform straight line fitting on the upper boundary line segment cluster, the lower boundary line segment cluster, the left boundary line segment cluster and the right boundary line segment cluster respectively, and obtain four target boundary lines; Obtain the intersection points of the four target boundary lines, and obtain the four vertex positions, thereby obtaining the second set of vertex positions. 4. The correction method according to claim 1, wherein the number of the target vertices is four, and the two-dimensional code image to be corrected is corrected based on the position of the target vertex to obtain a target two-dimensional code image include: determining the position of the target vertex on the two-dimensional code image to be corrected to determine the boundary point of the two-dimensional code image to be corrected; Perform perspective transformation on the to-be-corrected two-dimensional code image, correct the deformation of the to-be-corrected two-dimensional code image, and obtain a target two-dimensional code image. 5. The calibration method according to any one of claims 1 to 4, wherein the calibration method further comprises: Use the third algorithm to perform edge detection on the target two-dimensional code image to obtain a first edge contour map, and use the fourth algorithm to perform edge detection on the target two-dimensional code image to obtain a second edge contour map; Fusing the first edge contour map and the second edge contour map to obtain a target edge contour map; The target two-dimensional code image is sharpened through the target edge contour map to enhance the edge of the target two-dimensional code image. 6. correction method according to claim 5, is characterized in that, described 3rd algorithm is Canny edge detection algorithm, and described 4th algorithm is Sobel edge detection algorithm; The first edge contour map and the second edge contour map are fused to obtain the target edge contour map including: The first edge contour map and the second edge contour map are pixel superimposed according to a preset ratio to obtain a target edge contour. 7 . The correction method according to claim 1 , wherein the processing of the two-dimensional code image to be corrected to obtain the boundary contour of the two-dimensional code image to be corrected comprises: 8 . Binarize the QR code image to be corrected; Morphological corrosion is performed on the two-dimensional code image to be corrected that has undergone the binarization process to obtain the boundary contour of the two-dimensional code image to be corrected. 8. A correction device, comprising at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, The instructions are programmed to perform a correction method as claimed in any one of claims 1-7.

Images