CN108960259A - A kind of license plate preprocess method based on HSV - Google Patents

Abstract

A kind of license plate preprocess method based on HSV, belongs to field of intelligent transportation technology.It will obtain license plate image from RGB color and be converted to hsv color space, obtain I by handling HSV triple channel_HSV, by I_HSVTone, saturation degree and brightness handle respectively, then by H_i,j、S^* _i,jAnd V^* _i,jTriple channel carries out channel fusion, and image is rotated back into RGB color from hsv color space, obtains new RGB image I, and new images successively carry out gray processing, 3*3 median filtering, OTSU binaryzation after carrying out horizontal tilt correction, obtain pretreated license plate.The license plate preprocess method based on HSV that the present invention is obtained by using above-mentioned technology, it can effectively weaken the influence to license plate such as illumination, noise, shade, stain, and obtained bianry image is easy to carry out subsequent processing.

Description

A Preprocessing Method of License Plate Based on HSV

technical field

The invention relates to the technical field of intelligent transportation, in particular to an HSV-based license plate preprocessing method, which reduces the influence of light, noise, shadows, stains, etc. on the license plate binarization by processing the HSV three channels.

Background technique

With the continuous development of my country's transportation network, security issues have become increasingly prominent. The license plate is the identification of the vehicle, but in real life, due to the influence of the scene or human factors, this identification is often not easy to be recognized. In smart transportation projects, the license plate is often preprocessed first, then the characters are segmented on the license plate binary image, and finally the most segmented characters are recognized. Therefore, obtaining an easy-to-handle binary image is a key to the automatic license plate recognition system.

Commonly used license plate preprocessing includes the following steps: image normalization, histogram equalization, grayscale stretching, image smoothing, skew correction, and binarization. Li Zhanming (Research on License Plate Image Preprocessing in License Plate Number Recognition System [J]. Science Technology and Engineering, 2008, 8(8): 2081-2084.), Dong Lingjiao (Research on License Plate Image Preprocessing [J]. Mechanical and Electrical Engineering, 2009 ,26(6):107-109.) and Chen Ruozhu (Research on Image Preprocessing Based on License Plate Recognition [J]. Industrial Instrumentation and Automation Devices, 2015(4):40-43.) introduced related technologies. However, general license plate preprocessing steps are not robust to special situations in practical applications such as uneven illumination, license plate defacement, etc.

Compared with general license plate preprocessing operations, more complex preprocessing steps include: uneven illumination treatment, defacement repair, etc. For example, Zhao Kun (an improved preprocessing method for license plate recognition [J]. Henan Science, 2010, 28(3): 000329-332.) proposes to use local thresholds for binarization after light space correction combined with homomorphic filtering , to obtain an ideal binary image, but this method only realizes the processing of excessive local illumination of the license plate, for the shadow. Another example is Ma Chaoyu (A license plate image binarization algorithm that integrates multi-level light processing [J]. Computer Applications, 2013, 33 (a02): 200-202.) Based on the license plate grayscale image obtained by weighting, it is smoothed with the top hat algorithm The background light is smoothed by the Retinex algorithm, and the final binary image of the license plate is obtained by the local threshold method. Although this method can uniformly illuminate, its computational complexity is high. The license plate shadow problem is a sub-problem of the uneven illumination problem. Xu Yunjing et al. (A binarization method for shadowed license plate images, CN106650728A[P].2017) used the red channel image of the original license plate as the grayscale image, based on finding the bright and dark points Boundary realizes the binarization of license plate images with regular shadows, but this method cannot deal with irregular shadows. Bu Ke et al. (A preprocessing method for license plate images, CN102509095A[P].2012.) firstly obtain the upper, lower, left, and right boundaries of the license plate, and then find the Yin-Yang dividing line (that is, the light-dark dividing line between the shaded area and the non-shaded area) to convert the license plate It is divided into upper and lower parts, and the gray thresholds of the upper and lower parts are calculated respectively, and finally the gray level of the pixels in the upper part is supplemented. This method requires that the license plate obtained by positioning has clear upper, lower, left, and right boundaries, and can only deal with regular shadows. The repair of license plate stains refers to grasping the integrity of the license plate, interpolating the stains based on neighborhood pixels or directly replacing them with neighbor pixels. The computational complexity of such methods is very high.

To sum up, the shortcomings of the current license plate preprocessing operation include: 1) the general preprocessing steps are simple and efficient, but cannot deal with unevenly illuminated license plates and defaced license plates; 2) for unevenly illuminated license plates, there are only Regular shadows and algorithms are too time-consuming; 3) The complexity of the defacement repair algorithm is too high, and it cannot be applied to the preprocessing link of real-time license plate recognition for the time being.

Contents of the invention

For the above-mentioned problems existing in the prior art, the object of the present invention is to provide a kind of license plate preprocessing method based on HSV, it is by processing HSV three channels, weakens illumination, noise, shadow, stain etc. to the license plate binarization The impact of the method overcomes the shortcomings of the general license plate image preprocessing method, and obtains a binary image that is easy to carry out subsequent processing.

Described a kind of license plate preprocessing method based on HSV is characterized in that comprising the steps:

Step 1: Obtain the license plate image I _RGB , wherein the height of the image is h, the width is w, and the unit is pixel;

Step 2: Convert the license plate image from RGB color space to HSV color space to obtain I _HSV , and record the three channels of hue, saturation and brightness of I _HSV as H _i,j , S _i,j and V _i,j respectively, where i and j represent the abscissa and ordinate of the image pixel respectively, and 1≤i≤h, 1≤j≤w;

Step 3: For the brightness V _i,j in step 2), use the Hough line detection method to determine the horizontal tilt angle α of the image;

Step 4: For the saturation S _i,j in step 2), according to the formulas (1)~(4), use the Sauvola local threshold binarization method to achieve binarization, as follows: first obtain each pixel point ( i, j) the adaptive threshold T(i, j) in its c*c neighborhood range, and then realize binarization according to formula (4), and obtain S ^* _{i, j} ;

Among them, c represents the range of the pixel neighborhood, m(i,j) represents the mean value of S _i,j in the c*c neighborhood, σ(i,j) represents the mean square error of Si _,j in the c*c neighborhood, k ₁ represents the weight factor of the mean square error, and R ₁ represents the floating range of the mean square error change;

Step 5: For the luminance V _i,j in step 2), according to formulas (5)-(8), use the Sauvola local threshold value binarization method to realize binarization, as follows: first obtain each pixel point (i , j) the adaptive threshold T ^* (i, j) in its c*c neighborhood range, then realize binarization by formula (8), obtain V ^* _{i, j} ;

Among them, m ^* (i, j) represents the mean value of V _{i, j} in the c*c neighborhood, σ ^* (i, j) represents the mean square error of V _{i, j} in the c*c neighborhood, k ₂ means the mean square error The weight factor of , R ₂ represents the floating range of mean square error change;

Step 6: Perform channel fusion on the three channels of H _i,j , S ^* _i,j and V ^* _i,j , and convert the image from the HSV color space back to the RGB color space to obtain a new RGB image I;

Step 7: For the RGB image I in step 6), use the α calculated in step 3) to carry out horizontal tilt correction on the image to obtain image M;

Step 8: For the image M obtained in step 7), perform grayscale conversion, 3*3 median filtering, and OTSU binarization in sequence to obtain the preprocessing result G.

The advantages of the present invention are: by adopting the above-mentioned technology, the present invention can effectively reduce the influence of illumination, noise, shadow, stain, etc. on the license plate, and the obtained binary image is easy to carry out subsequent processing.

Description of drawings

Fig. 1 is the decolorization effect figure of five license plate images that the embodiment of the present invention chooses;

Fig. 2 is five license plate image decolorization effect figures after step 7 processing of the present invention;

Fig. 3 is five license plate images after step 8 processing of the present invention.

Detailed ways

The specific implementation of the HSV-based license plate preprocessing method of the present invention will be described in detail below in conjunction with the examples.

As shown in the figure, the license plate preprocessing method based on HSV of the present invention comprises the following steps:

Step 1: Obtain the license plate image I _RGB , as shown in Figure 1, wherein the height of the image is h, the width is w, and the unit is pixel;

Step 2: Convert the license plate image from RGB color space to HSV color space to obtain I _HSV , and record the three channels of hue, saturation and brightness of I _HSV as H _i,j , S _i,j and V _i,j respectively, where i and j represent the abscissa and ordinate of the image pixel respectively, and 1≤i≤h, 1≤j≤w;

Step 3: For the brightness V _i,j in step 2), use the Hough line detection method to determine the horizontal tilt angle α of the image;

Step 4: For the saturation S _i,j in step 2), according to the formulas (1)~(4), use the Sauvola local threshold binarization method to achieve binarization, as follows: first obtain each pixel point ( i, j) the adaptive threshold T(i, j) in its c*c neighborhood range, and then realize binarization according to formula (4), and obtain S ^* _{i, j} ;

Among them, c represents the range of the pixel neighborhood, m(i,j) represents the mean value of S _i,j in the c*c neighborhood, σ(i,j) represents the mean square error of Si _,j in the c*c neighborhood, k ₁ represents the weight factor of the mean square error, and R ₁ represents the floating range of the mean square error change; in this example, c=15, k=0.5, R=128;

Step 5: For the luminance V _i,j in step 2), according to formulas (5)-(8), use the Sauvola local threshold value binarization method to realize binarization, as follows: first obtain each pixel point (i , j) the adaptive threshold T ^* (i, j) in its c*c neighborhood range, then realize binarization by formula (8), obtain V ^* _{i, j} ;

Among them, m ^* (i, j) represents the mean value of V _{i, j} in the c*c neighborhood, σ ^* (i, j) represents the mean square error of V _{i, j} in the c*c neighborhood, k ₂ means the mean square error The weight factor of , R ₂ represents the floating range of mean square error change; in this example, k ₂ =0.5, R ₂ =128;

Step 6: Perform channel fusion on the three channels of H _i,j , S ^* _i,j and V ^* _i,j , and convert the image from the HSV color space back to the RGB color space to obtain a new RGB image I;

Step 7: For the RGB image I in step 6, use the α calculated in step 3 to correct the horizontal tilt of the image to obtain the image M, as shown in Figure 2;

Step 8: For the image M in step 7, perform grayscale conversion, 3*3 median filter, and OTSU binarization in sequence to obtain the preprocessing result G, as shown in Figure 3.

The content described in the embodiments of this specification is only an enumeration of the implementation forms of the inventive concept. The protection scope of the present invention should not be regarded as limited to the specific forms stated in the embodiments. The protection scope of the present invention also extends to the field Equivalent technical means that the skilled person can think of based on the concept of the present invention.

Claims (1)

1. a license plate preprocessing method based on HSV, is characterized in that comprising the steps: Step 1: Obtain the license plate image I _RGB , wherein the height of the image is h, the width is w, and the unit is pixel; Step 2: Convert the license plate image from RGB color space to HSV color space to obtain I _HSV , and record the three channels of hue, saturation and brightness of I _HSV as H _i,j , S _i,j and V _i,j respectively, where i and j represent the abscissa and ordinate of the image pixel respectively, and 1≤i≤h, 1≤j≤w; Step 3: For the brightness V _i,j in step 2), use the Hough line detection method to determine the horizontal tilt angle α of the image; Step 4: For the saturation S _i,j in step 2), according to the formulas (1)~(4), use the Sauvola local threshold binarization method to achieve binarization, as follows: first obtain each pixel point ( i, j) the adaptive threshold T(i, j) in its c*c neighborhood range, and then realize binarization according to formula (4), and obtain S ^* _{i, j} ; Among them, c represents the range of the pixel neighborhood, m(i,j) represents the mean value of S _i,j in the c*c neighborhood, σ(i,j) represents the mean square error of Si _,j in the c*c neighborhood, k ₁ represents the weight factor of the mean square error, and R ₁ represents the floating range of the mean square error change; Step 5: For the luminance V _i,j in step 2), according to formulas (5)-(8), use the Sauvola local threshold value binarization method to realize binarization, as follows: first obtain each pixel point (i , j) the adaptive threshold T ^* (i, j) in its c*c neighborhood range, then realize binarization by formula (8), obtain V ^* _{i, j} ; Among them, m ^* (i, j) represents the mean value of V _{i, j} in the c*c neighborhood, σ ^* (i, j) represents the mean square error of V _{i, j} in the c*c neighborhood, k ₂ means the mean square error The weight factor of , R ₂ represents the floating range of mean square error change; Step 6: Perform channel fusion on the three channels of H _i,j , S ^* _i,j and V ^* _i,j , and convert the image from the HSV color space back to the RGB color space to obtain a new RGB image I; Step 7: For the RGB image I in step 6), use the α calculated in step 3) to carry out horizontal tilt correction on the image to obtain image M; Step 8: For the image M obtained in step 7), perform grayscale conversion, 3*3 median filtering, and OTSU binarization in sequence to obtain the preprocessing result G.