CN110598589A

CN110598589A - Image pyramid-based palm print identification method, system, device and medium

Info

Publication number: CN110598589A
Application number: CN201910798336.XA
Authority: CN
Inventors: 刘欣; 冯先成; 龚威威; 刘莎
Original assignee: Wuhan Institute of Technology
Current assignee: Wuhan Institute of Technology
Priority date: 2019-08-27
Filing date: 2019-08-27
Publication date: 2019-12-20

Abstract

The invention relates to a palm print identification method, a system, a device and a medium based on an image pyramid, wherein the method comprises the steps of obtaining an original palm print image of a palm print to be matched; respectively preprocessing each original template palm print image and each original template palm print image to respectively obtain a target template palm print image set and a target palm print image; respectively extracting a target SIFT feature point set and a template feature point set of each target template palm print image by adopting a scale invariant feature transformation method; calculating to obtain the offset between the target palm print image and each target template palm print image according to each target template palm print image, each template feature point set, the target palm print image and the target SIFT feature point set based on an image pyramid method; and in the target template palm print image set, identifying and matching the palm prints to be matched according to all the offsets to obtain and output an identification result. The invention overcomes the influence of palm print deformation, translation and noise on palm print identification, and obviously improves the identification precision of the palm print.

Description

Image pyramid-based palm print identification method, system, device and medium

Technical Field

The invention relates to the technical field of biological feature recognition, in particular to a palm print recognition method, a palm print recognition system, a palm print recognition device and a palm print recognition medium based on an image pyramid.

Background

The palm print recognition technology has great development potential and application prospect as an important aspect of the biological feature recognition technology. Palm print recognition technology has evolved over two decades since 1997, due to its own advantages such as: compared with iris recognition, the method is easy to collect, has no harm to human body and is low in collection equipment; compared with human face recognition, the method can identify the twins more easily; compared with fingerprint identification, the method has higher identification precision and can adopt a non-contact acquisition mode; handwriting and gait are easily imitated and the biological characteristics are easily changed compared to handwriting and gait recognition, and the stability of palm print recognition is high. Therefore, palm print recognition is becoming one of the hot research directions in the field of biometric identification, and products related to the palm print recognition are also gradually advancing.

The stable and reliable information of the palm print, such as the texture characteristics of the palm print main line, the palm print folds, the palm print lines and the like, can realize reliable identity recognition. The texture feature structure of the palm print is unique, the structure also enables the anti-noise capability in the algorithm processing process to be outstanding, and even if an image with poor quality or low resolution is directly input, the main feature of the palm print can be relatively stably extracted.

Compared with traditional biometric identification, palm print identification has many natural advantages in terms of universality, distinctiveness, invariance, easy acquisition, acceptability, hardware cost and the like, and the advantages are favorable conditions for the future development of palm print identification.

However, in the technology of palm print identification, palm print deformation, noise and palm print blur are some key difficulties affecting palm print identification effect during the identification process of palm print with low resolution. One of the existing methods for solving palm print deformation is to extract feature descriptors of local areas corresponding to two palm prints and then match and recognize the palm prints by combining a matching method based on coding, but the scheme ignores the influence of palm print translation and noise on the palm print recognition method, so that the matching and recognition of the palm prints are still not accurate enough, the recognition precision is not high, and the recognition effect is not good.

Disclosure of Invention

The technical problem to be solved by the present invention is to provide a method, a system, a device and a medium for identifying a palm print based on an image pyramid, which overcome the influence of palm print translation and noise on palm print identification and obviously improve the identification precision of the palm print.

The technical scheme for solving the technical problems is as follows:

a palm print identification method based on an image pyramid comprises the following steps:

step 1: acquiring an original palm print image of a palm print to be matched;

step 2: respectively preprocessing each original template palm print image in a preset original template palm print image set to obtain a target template palm print image set, and preprocessing the original palm print images to obtain a target palm print image;

and step 3: respectively extracting a target SIFT feature point set of the target palm print image and a template feature point set corresponding to each target template palm print image in the target template palm print image set one by adopting a scale invariant feature transformation method;

and 4, step 4: calculating to obtain the offset between each target template palm print image and each target template palm print image according to each target template palm print image, the template feature point set corresponding to each target template palm print image one by one, the target palm print image and the target SIFT feature point set corresponding to the target palm print image based on an image pyramid method;

and 5: and in the target template palm print image set, identifying and matching the palm prints to be matched according to all the offsets to obtain and output an identification result.

The invention has the beneficial effects that: because the palm print image is easily influenced by noise, before extracting a target SIFT feature point set of a target palm print image and a template feature point set corresponding to each target template palm print image one by one, preprocessing each original template palm print image respectively to obtain one-to-one corresponding target template palm print image, wherein the target template palm print images form a target template palm print image set, and preprocessing the original palm print images to obtain a target palm print image;

the Scale-invariant Feature Transform (SIFT) method is an algorithm for detecting local features, and the method searches extreme points in a Scale space, extracts the position, Scale and rotation invariants of the extreme points, keeps the rotation, Scale scaling and brightness change unchanged, and keeps certain stability on perspective change, affine transformation and noise, so that a target SIFT Feature point set extracted by the SIFT method and a template Feature point set corresponding to each target template palm print image one by one are further beneficial to subsequent palm print identification and matching, the influence of palm print translation on palm print identification is overcome, and the identification precision of the palm prints is obviously improved;

the image pyramid is an effective but simple-concept structure for explaining the image with multiple resolutions, and the pyramid of one image is a series of image sets with gradually reduced resolutions arranged in a pyramid shape; after extracting the target SIFT feature point set and the template feature point set corresponding to each target template palm print image one by one, the method adopts an image pyramid method, can effectively correct palm print deformation in a local area, particularly effectively overcomes the influence of the palm print deformation on palm print identification in a low-resolution palm print identification process, is flexible and simple, is beneficial to subsequently calculating the offset between the target palm print image and each target template palm print image, and further improves the identification precision of the palm print;

the offset between the characteristic points of the same palm print is small, and the offset between the characteristic points of different palm prints is large, so that the target palm print image and the target template palm print image in the target template palm print image set can be judged to be most matched through the offsets, the palm print to be matched is identified, the identification matching method is simple and effective, and the identification precision of the palm print is obviously improved.

On the basis of the technical scheme, the invention can be further improved as follows:

further: in the step 2, the specific step of obtaining the target template palm print image set includes:

step 2 a.1: respectively extracting the interested region of each original template palm print image in the original template palm print image set to obtain a template palm print ROI image set;

step 2 a.2: respectively filtering each template palm print ROI image in the template palm print ROI image set by adopting an MFRAT filtering method to obtain a template palm print filtering image set;

step 2 a.3: in each template palm print filtering image of the template palm print filtering image set, respectively encoding each pixel point to obtain the target template palm print image set;

in the step 2, the specific step of obtaining the target palm print image includes:

step 2 b.1: extracting an interested area of the original palm print image to obtain a palm print ROI image;

step 2 b.2: filtering the palm print ROI image by adopting an MFRAT filtering method to obtain a palm print filtering image;

step 2 b.3: and respectively coding each pixel point in the palm print filtering image to obtain the target palm print image.

Further: the step 2a.2 specifically comprises:

step 2 a.2.1: respectively carrying out histogram equalization processing on each template palm print ROI image in the template palm print ROI image set to obtain a first intermediate template palm print image set;

step 2 a.2.2: respectively carrying out normalization processing on each first intermediate template palm print image in the first intermediate template palm print image set to obtain a second intermediate template palm print image set;

step 2 a.2.3: constructing a first MFRAT filter function in each second intermediate template palm print image of the second intermediate template palm print image set, establishing a p multiplied by p first filter grid by taking any pixel point as a first central point, calculating and obtaining a plurality of first response values of the first central point in each direction according to the first MFRAT filter function in the first filter grid, and obtaining a first pixel accumulated value of the first central point in each direction according to all the first response values in each direction;

the first center point (x, y) is at theta_kThe first MFRAT filter function in the direction is:

wherein, (x, y) is the coordinate of the first central point in the second intermediate template palm print image, r (x, y) is the pixel value corresponding to the first central point (x, y), and θ_k(k-1, 2, …,6) is selected from six directions of 0, pi/6, 2 pi/6, 3 pi/6, 4 pi/6 and 5 pi/6,for the first filter grid at θ_kAn equation of a straight line in the direction,is the first center point (x, y) at theta_kIn the direction,A first response value of the linear equation;

step 2 a.2.4: traversing each pixel point of each second intermediate template palm print image, obtaining a first pixel accumulated value of each pixel point in each direction according to the method of the step 2a.2.3, and obtaining a template palm print filtering image corresponding to one corresponding second intermediate template palm print image according to all the first pixel accumulated values of all the pixel points;

step 2 a.2.5: obtaining a template palmprint filtering image set according to all template palmprint filtering images;

the step 2b.2 specifically comprises:

step 2 b.2.1: performing histogram equalization processing on the palm print ROI image to obtain a first middle palm print image;

step 2 b.2.2: normalizing the first intermediate palm print image to obtain a second intermediate palm print image;

step 2 b.2.3: constructing a second MFRAT filter function in the second intermediate palm print image, establishing a p × p second filter grid by taking any pixel point as the pixel point, calculating and obtaining a plurality of second response values of the second center point in each direction in the second filter grid according to the second MFRAT filter function, and obtaining a second pixel accumulated value of the second center point in each direction according to all the second response values in each direction;

the second center point (x ', y') is at theta_kThe second MFRAT filter function in the direction is:

wherein, (x ', y ') is the coordinate of the second central point in the second intermediate palm print image, r ' (x ', y ') is the pixel value corresponding to the second central point (x ', y '),for the second filter grid at θ_kAn equation of a straight line in the direction,is the second central point (x)', y') at θ_kIn the direction,A second response value of the linear equation;

step 2 b.2.4: traversing each pixel point of the second middle palm print image, and obtaining a second pixel accumulated value of each pixel point in each direction according to the method of the step 2 b.2.3;

step 2 b.2.5: and obtaining the palm print filtering image according to all the second pixel accumulated values of all the pixel points.

Further: the step 2a.3 specifically comprises:

step 2 a.3.1: in each template palm print filtering image in the template palm print filtering image set, taking the direction corresponding to the maximum value in all the first pixel accumulated values of each pixel point as the first characteristic coding value of the corresponding pixel point;

step 2 a.3.2: obtaining a first feature coding value subset corresponding to each template palm print filtering image one by one according to all first feature coding values corresponding to all pixel points in each template palm print filtering image;

step 2 a.3.3: obtaining the target template palm print image set according to all the first feature coding value subsets corresponding to all the template palm print filtering images;

the step 2b.3 specifically comprises:

step 2 b.3.1: in the palm print filtering image, taking the direction corresponding to the maximum value in all the second pixel accumulated values of each pixel point as a second characteristic coding value of the corresponding pixel point;

step 2 b.3.2: obtaining a second feature code value subset corresponding to the palm print filtering image according to all second feature code values corresponding to all pixel points in the palm print filtering image;

step 2 b.3.3: and obtaining the target palm print image according to the second characteristic coding value subset.

Further: in the step 3, the specific step of extracting the template feature point set corresponding to each target template palm print image in the target template palm print image set one to one includes:

step 3 a.1: respectively expanding each target template palm print image by adopting a bilinear interpolation method to obtain an expanded target template palm print image set;

step 3 a.2: in each enlarged target template palm print image of the enlarged target template palm print image set, adopting a scale invariant feature transformation method to construct a first scale space, and detecting a first extreme point set corresponding to each enlarged target template palm print image in the first scale space one by one according to a preset first pixel threshold;

step 3 a.3: filtering a first extreme point set of each enlarged target template palm print image in the first scale space by adopting a Harris Commer detector to obtain a template feature point set corresponding to each target template palm print image one to one;

in the step 3, the specific step of extracting the target SIFT feature point set of the target palm print image includes:

step 3 b.1: expanding the target palm print image by adopting a bilinear interpolation method to obtain an expanded target palm print image;

step 3 b.2: constructing a second scale space in the enlarged target palm print image set by adopting a scale-invariant feature transformation method, and detecting a corresponding second extreme point set of the enlarged target palm print image in the second scale space according to a preset second pixel threshold;

step 3 b.3: and filtering all second extreme points in the second extreme point set by adopting a Harris comber detector to obtain the target SIFT feature point set corresponding to the target palm print image.

Further: the specific steps of the step 4 comprise:

step 4.1: acquiring a template corner point coordinate set corresponding to each target template palm print image one by one according to each target template palm print image and a corresponding template feature point set based on an image pyramid method, and acquiring a target corner point coordinate set corresponding to the target palm print image according to the target palm print image and the target SIFT feature point set;

step 4.2: and respectively calculating the offset between the target palm print image and each target template palm print image according to each template corner point coordinate set and the target corner point coordinate set by adopting a BLPOC (global binary offset plus one minus one plus.

Further: the concrete implementation of the step 5 is as follows:

and taking the target template palm print image corresponding to the minimum value in all the offsets as the identification result of the palm print to be matched and outputting the identification result.

According to another aspect of the present invention, there is provided an image pyramid-based palmprint recognition system, including an obtaining module, a preprocessing module, an extracting module, a calculating module, and a recognition module:

the acquisition module is used for acquiring an original palm print image of a palm print to be matched;

the preprocessing module is used for respectively preprocessing each original template palm print image in a preset original template palm print image set to obtain a target template palm print image set, and is also used for preprocessing the original palm print images to obtain a target palm print image;

the extraction module is used for respectively extracting a target SIFT feature point set of the target palm print image and a template feature point set corresponding to each target template palm print image in the target template palm print image set by adopting a scale invariant feature transformation method;

the calculation module is used for calculating to obtain offset information between the target SIFT feature point set and the template feature point set corresponding to each target template palm print image one by one based on an image pyramid method;

and the identification module is used for identifying and matching the palm print to be matched according to all the offset information in the target template palm print image set to obtain an identification result and outputting the identification result.

The invention has the beneficial effects that: the method comprises the steps that original palm print images of palm prints to be matched are obtained through an obtaining module, then each original template palm print image is preprocessed through a preprocessing module respectively to obtain target template palm print images which correspond to one another, the target template palm print images form a target template palm print image set, the original palm print images are preprocessed in the same way to obtain target palm print images, through preprocessing, palm print line features affected by noise can be filtered, the noise resistance of a follow-up scale invariant feature transformation method is improved, the influence of the noise on palm print recognition is effectively overcome, meanwhile, the contrast of texture features of each target template palm print image and the target palm print images is enhanced, and follow-up palm print recognition and matching are facilitated; the target SIFT feature point set extracted by the SIFT method through the extraction module and the template feature point set corresponding to each target template palm print image one by one are further beneficial to subsequent palm print identification and matching, the influence of palm print translation on palm print identification is overcome, and the identification precision of the palm print is obviously improved; after extracting the template feature point sets corresponding to the target SIFT feature point sets and each target template palm print image one by one, the image pyramid method is adopted by the calculation module, palm print deformation can be effectively corrected in a local area, especially for the palm print recognition process with low resolution, the influence of the palm print deformation on palm print recognition is effectively overcome, the method is flexible and simple, the offset between the target palm print image and each target template palm print image can be favorably calculated subsequently, and the recognition precision of the palm print is further improved; and finally, the recognition module can judge which target template palm print image in the target template palm print image set is most matched with the target palm print image through the offset, so that the palm print to be matched is recognized, the recognition matching method is simple and effective, and the recognition precision of the palm print is obviously improved.

further: the preprocessing module comprises an interested region extracting unit, a filtering unit and a coding unit;

the interesting region extracting unit is used for respectively extracting the interesting region of each original template palm print image in the original template palm print image set to obtain a template palm print ROI image set; the method is also used for extracting an interested area of the original palm print image to obtain a palm print ROI image;

the filtering unit is used for respectively filtering each template palm print ROI image in the template palm print ROI image set by adopting an MFRAT filtering method to obtain a template palm print filtering image set; the palm print ROI image is filtered by adopting an MFRAT filtering method to obtain a palm print filtering image;

the encoding unit is used for respectively encoding each pixel point in each template palm print filtering image of the template palm print filtering image set to obtain the target template palm print image set; and the method is also used for respectively coding each pixel point in the palm print filtering image to obtain the target palm print image.

Further: the filtering unit is specifically configured to:

respectively carrying out histogram equalization processing on each template palm print ROI image in the template palm print ROI image set to obtain a first intermediate template palm print image set;

respectively carrying out normalization processing on each first intermediate template palm print image in the first intermediate template palm print image set to obtain a second intermediate template palm print image set;

in each second intermediate template palm print image of the second intermediate template palm print image set, constructing a first MFRAT filter function, establishing a p × p first filter grid by taking any pixel point as a first central point, calculating to obtain a plurality of first response values of the first central point in each direction according to the first MFRAT filter function in the first filter grid, and obtaining a first pixel accumulated value of the first central point in each direction according to all the first response values in each direction;

wherein, (x, y) is the coordinate of the first central point in the second intermediate template palm print image, r (x, y) is the pixel value corresponding to the first central point (x, y), and θ_k(k-1, 2, …,6) is selected six directions0, pi/6, 2 pi/6, 3 pi/6, 4 pi/6 and 5 pi/6 respectively,for the first filter grid at theta_kAn equation of a straight line in the direction,is the first center point (x, y) at theta_kEquation of straight line in directionA first response value of;

traversing each pixel point of each second intermediate template palm print image, obtaining a first pixel accumulated value of each pixel point in each direction according to the method of the step 2a.2.3, and obtaining a template palm print filtering image corresponding to one corresponding second intermediate template palm print image according to all the first pixel accumulated values of all the pixel points;

obtaining a template palmprint filtering image set according to all template palmprint filtering images;

the filtering unit is further specifically configured to:

performing histogram equalization processing on the palm print ROI image to obtain a first middle palm print image;

normalizing the first intermediate palm print image to obtain a second intermediate palm print image;

in the second intermediate palm print image, constructing a second MFRAT filtering function, establishing a p × p second filtering grid by taking any pixel point as a second central point, calculating to obtain a plurality of second response values of the second central point in each direction according to the second MFRAT filtering function in the second filtering grid, and obtaining a second pixel accumulated value of the second central point in each direction according to all the second response values in each direction;

wherein, (x ', y ') is the coordinate of the pixel point of the second central point in the second intermediate palm print image, r ' (x ', y ') is the pixel value corresponding to the second central point (x ', y '),for the second filter grid at theta_kAn equation of a straight line in the direction,is the second center point (x ', y') is at theta_kEquation of straight line in directionA second response value of;

traversing each pixel point of the second middle palm print image, and obtaining a second pixel accumulated value of each pixel point in each direction according to the method of the step 2 b.2.3;

and obtaining the palm print filtering image according to all the second pixel accumulated values of all the pixel points.

Further: the encoding unit is specifically configured to:

in each template palm print filtering image in the template palm print filtering image set, taking the direction corresponding to the maximum value in all the first pixel accumulated values of each pixel point as the first characteristic coding value of the corresponding pixel point;

obtaining a first feature coding value subset corresponding to each template palm print filtering image one by one according to all first feature coding values corresponding to all pixel points in each template palm print filtering image;

obtaining the target template palm print image set according to all the first feature coding value subsets corresponding to all the template palm print filtering images;

the encoding unit is further specifically configured to:

in the palm print filtering image, taking the direction corresponding to the maximum value in all the second pixel accumulated values of each pixel point as a second characteristic coding value of the corresponding pixel point;

obtaining a second feature code value subset corresponding to the palm print filtering image according to all second feature code values corresponding to all pixel points in the palm print filtering image;

and obtaining the target palm print image according to the second characteristic coding value subset.

Further: the extraction module is specifically configured to:

respectively expanding each target template palm print image by adopting a bilinear interpolation method to obtain an expanded target template palm print image set;

in each enlarged target template palm print image of the enlarged target template palm print image set, adopting a scale invariant feature transformation method to construct a first scale space, and detecting a first extreme point set corresponding to each enlarged target template palm print image in the first scale space one by one according to a preset first pixel threshold;

filtering a first extreme point set of each enlarged target template palm print image in the first scale space by adopting a Harris Commer detector to obtain a template feature point set corresponding to each target template palm print image one to one;

the extraction module is further specifically configured to:

expanding the target palm print image by adopting a bilinear interpolation method to obtain an expanded target palm print image;

constructing a second scale space in the enlarged target palm print image set by adopting a scale-invariant feature transformation method, and detecting a corresponding second extreme point set of the enlarged target palm print image in the second scale space according to a preset second pixel threshold;

and filtering all second extreme points in the second extreme point set by adopting a Harris comber detector to obtain the target SIFT feature point set corresponding to the target palm print image.

Further: the calculation module is specifically configured to:

acquiring a template corner point coordinate set corresponding to each target template palm print image one by one according to each target template palm print image and a corresponding template feature point set based on an image pyramid method, and acquiring a target corner point coordinate set corresponding to the target palm print image according to the target palm print image and the target SIFT feature point set;

and respectively calculating the offset between the target palm print image and each target template palm print image according to each template corner point coordinate set and the target corner point coordinate set by adopting a BLPOC (global binary offset plus one minus one plus.

Further: the identification module is specifically configured to:

According to another aspect of the present invention, an image pyramid-based palm print recognition apparatus is provided, which includes a processor, a memory, and a computer program stored in the memory and executable on the processor, where the computer program implements the steps in an image pyramid-based palm print recognition method of the present invention when running.

The invention has the beneficial effects that: the palm print recognition method is realized by the computer program stored in the memory and running on the processor, and the palm print deformation is effectively corrected based on the scale invariant feature transformation method and the image pyramid, so that the influence of the palm print deformation, the palm print translation and noise on the palm print recognition is overcome, the recognition precision of the palm print is obviously improved, and the recognition effect is good.

In accordance with another aspect of the present invention, there is provided a computer storage medium comprising: at least one instruction, when executed, implementing the steps of the image pyramid-based palm print identification method of the present invention.

The invention has the beneficial effects that: the palm print recognition method is realized by executing the computer storage medium containing at least one instruction, and the palm print deformation is effectively corrected based on the scale invariant feature transformation method and the image pyramid, so that the influence of the palm print deformation, palm print translation and noise on the palm print recognition is overcome, the palm print recognition precision is obviously improved, and the recognition effect is good.

Drawings

Fig. 1 is a schematic flowchart of a palm print recognition method based on an image pyramid according to an embodiment of the present invention;

FIG. 2 is a schematic flow chart illustrating a process of obtaining a target template palm print image set according to a first embodiment of the present invention;

FIG. 3 is a schematic flow chart illustrating a process of obtaining a target palm print image according to a first embodiment of the present invention;

FIG. 4 is a schematic flow chart illustrating a process of obtaining a template feature point set according to a first embodiment of the present invention;

fig. 5 is a schematic flow chart illustrating a process of obtaining a target SIFT feature point set in the first embodiment of the present invention;

FIG. 6 is a flowchart illustrating a process of calculating an offset according to a first embodiment of the present invention;

fig. 7 is a schematic structural diagram of a palm print recognition system based on an image pyramid according to a second embodiment of the present invention;

fig. 8 is a second schematic structural diagram of a palm print recognition system based on an image pyramid in the second embodiment of the present invention.

Detailed Description

The principles and features of this invention are described below in conjunction with the following drawings, which are set forth by way of illustration only and are not intended to limit the scope of the invention.

The present invention will be described with reference to the accompanying drawings.

In an embodiment, as shown in fig. 1, a palm print recognition method based on an image pyramid includes the following steps:

s1: acquiring an original palm print image of a palm print to be matched;

s2: respectively preprocessing each original template palm print image in a preset original template palm print image set to obtain a target template palm print image set, and preprocessing the original palm print images to obtain a target palm print image;

s3: respectively extracting a target SIFT feature point set of the target palm print image and a template feature point set corresponding to each target template palm print image in the target template palm print image set one by adopting a scale invariant feature transformation method;

s4: calculating to obtain the offset between each target template palm print image and each target template palm print image according to each target template palm print image, the template feature point set corresponding to each target template palm print image one by one, the target palm print image and the target SIFT feature point set corresponding to the target palm print image based on an image pyramid method;

s5: and in the target template palm print image set, identifying and matching the palm prints to be matched according to all the offsets to obtain and output an identification result.

Firstly, preprocessing each original template palm print image respectively to obtain target template palm print images in one-to-one correspondence, wherein the target template palm print images form a target template palm print image set, and preprocessing the original palm print images to obtain target palm print images; the target SIFT feature point set extracted by the SIFT method and the template feature point set corresponding to each target template palm print image one by one are further beneficial to subsequent palm print identification and matching, the influence of palm print translation on palm print identification is overcome, and the identification precision of the palm print is obviously improved; after extracting the template feature point set corresponding to the target SIFT feature point set and each target template palm print image one by one, the image pyramid method is adopted, palm print deformation can be effectively corrected in a local area, the method is flexible and simple, the offset between the target palm print image and each target template palm print image can be calculated subsequently, the influence of palm print deformation on palm print identification is effectively overcome, and the identification precision of the palm print is further improved; the offset between the characteristic points of the same palm print is small, and the offset between the characteristic points of different palm prints is large, so that the target palm print image and the target template palm print image in the target template palm print image set can be judged to be most matched through the offsets, the palm print to be matched is identified, the identification matching method is simple and effective, and the identification precision of the palm print is obviously improved.

It should be understood that the original template palm print image set in the present invention includes a plurality of original template palm print images, which are collected in advance, for example, collected by web crawler or manual collection; therefore, the target template palm print image set contains a plurality of target template palm print images.

Preferably, as shown in fig. 2, in S2, the specific step of obtaining the target template palm print image set includes:

s2a.1: respectively extracting the interested region of each original template palm print image in the original template palm print image set to obtain a template palm print ROI image set;

s2a.2: respectively filtering each template palm print ROI image in the template palm print ROI image set by adopting an MFRAT filtering method to obtain a template palm print filtering image set;

s2a.3: in each template palm print filtering image of the template palm print filtering image set, respectively encoding each pixel point to obtain the target template palm print image set;

as shown in fig. 3, in S2, the specific step of obtaining the target palm print image includes:

s2b.1: extracting an interested area of the original palm print image to obtain a palm print ROI image;

s2b.2: filtering the palm print ROI image by adopting an MFRAT filtering method to obtain a palm print filtering image;

s2b.3: and respectively coding each pixel point in the palm print filtering image to obtain the target palm print image.

Because the region of Interest (ROI) contains more effective palm print features, the region of Interest of each original template palm print image in the original template palm print image set and the region of Interest of the original palm print image are extracted at first, so that the effective palm print features in each palm print image can be extracted, the subsequent feature extraction and palm print identification matching are facilitated, and the operation amount can be greatly reduced; the MFRAT (modified finish Radon transform) filtering method is an improved limited Radon transform, and can accurately position palm print characteristics, so that the influence of noise on palm print identification can be effectively avoided through the filtering treatment of the MFRAT filtering method, and each pixel point in each template palm print filtering image and each pixel point in the palm print filtering image are respectively encoded, so that the subsequent characteristic extraction and the execution of an image pyramid method are facilitated, and the identification and the matching of palm prints are facilitated.

Preferably, s2a.2 specifically comprises:

s2a.2.1: respectively carrying out histogram equalization processing on each template palm print ROI image in the template palm print ROI image set to obtain a first intermediate template palm print image set;

s2a.2.2: respectively carrying out normalization processing on each first intermediate template palm print image in the first intermediate template palm print image set to obtain a second intermediate template palm print image set;

s2a.2.3: in each second intermediate template palm print image of the second intermediate template palm print image set, constructing a first MFRAT filter function, establishing a p × p first filter grid by taking any pixel point as a first central point, calculating to obtain a plurality of first response values of the first central point in each direction according to the first MFRAT filter function in the first filter grid, and obtaining a first pixel accumulated value of the first central point in each direction according to all the first response values in each direction;

wherein, (x, y) is the coordinate of the first central point in the second intermediate template palm print image, r (x, y) is the pixel value corresponding to the first central point (x, y), and θ_k(k-1, 2, …,6) is selected from six directions of 0, pi/6, 2 pi/6, 3 pi/6, 4 pi/6 and 5 pi/6,for the first filter grid at theta_kAn equation of a straight line in the direction,is the first center point (x, y) at theta_kEquation of straight line in directionA first response value of;

s2a.2.4: traversing each pixel point of each second intermediate template palm print image, obtaining a first pixel accumulated value of each pixel point in each direction according to the method of the step 2a.2.3, and obtaining a template palm print filtering image corresponding to one corresponding second intermediate template palm print image according to all the first pixel accumulated values of all the pixel points;

s2a.2.5: obtaining a template palmprint filtering image set according to all template palmprint filtering images;

the S2b.2 specifically comprises the following steps:

s2b.2.1: performing histogram equalization processing on the palm print ROI image to obtain a first middle palm print image;

s2b.2.2: normalizing the first intermediate palm print image to obtain a second intermediate palm print image;

s2b.2.3: constructing a second MFRAT filter function in the second intermediate palm print image, establishing a second filter grid of p multiplied by p by taking any pixel point as a second central point, calculating and obtaining a plurality of second response values of the second central point in each direction in the second filter grid according to the MFRAT filter function, and obtaining a second pixel accumulated value of the second central point in each direction according to all the second response values in each direction;

wherein, (x ', y ') is the coordinate of the second central point in the second intermediate palm print image, r ' (x ', y ') is the pixel value corresponding to the second central point (x ', y '),for the second filter grid at theta_kAn equation of a straight line in the direction,is the second center point (x ', y') is at theta_kEquation of straight line in directionA second response value of;

s2b.2.4: traversing each pixel point of the second middle palm print image, and obtaining a second pixel accumulated value of each pixel point in each direction according to the method of the step 2 b.2.3;

s2b.2.5: and obtaining the palm print filtering image according to all the second pixel accumulated values of all the pixel points.

Through the filtering processing in the steps, on one hand, the influence of noise on palm print recognition can be effectively overcome, on the other hand, the contrast of palm print features in each image can be enhanced, so that the execution of a subsequent SIFT method and an image pyramid method is facilitated, an accurate target SIFT feature point set and a template feature point set are extracted, and the accurate offset between the target palm print image and each target template palm print image is calculated conveniently.

Specifically, one template palm print ROI image I (x, y) in the template palm print ROI image set of the present embodiment has a size of M × N (M and N are the total number of row pixel points and the total number of column pixel points of the template palm print ROI image, respectively); histogram equalization processing is carried out on the template palm print ROI image I (x, y), and a corresponding first intermediate template palm print ROI image I is obtained_H(x, y), and taking the first intermediate template palm print ROI image as I_H(x, y) carrying out normalization processing to obtain a normalized second intermediate template palmThe texture ROI image is I_N(x, y); and finally, constructing a 11 multiplied by 11 first filter grid by taking the pixel point (x, y) as a first central point, filtering in the first filter grid according to a first MFRAT filter function, and calculating to obtain a first accumulated pixel value M of the first central point in six directions respectively_iWherein i is 1,2, …, 6; traversing the second intermediate template palmprint ROI image as I_NCalculating the first pixel accumulated value corresponding to each pixel point one by one according to the same method for each pixel point in (x, y), and obtaining the second middle template palm print ROI image as I according to all the first pixel accumulated values_NThe (x, y) corresponding template palm print filtering image in six directions is I_Mi(x, y), wherein i ═ 1,2, …, 6; according to the same method, template palm print filtering images in six directions corresponding to each template palm print ROI image of the template palm print ROI image set and a palm print filtering image J corresponding to the palm print ROI image can be obtained_Mi(x ', y'), wherein i ═ 1,2, …, 6.

Preferably, s2a.3 specifically comprises:

s2a.3.1: in each template palm print filtering image in the template palm print filtering image set, taking the direction corresponding to the maximum value in all the first pixel accumulated values of each pixel point as the first characteristic coding value of the corresponding pixel point;

s2a.3.2: obtaining a first feature coding value subset corresponding to each template palm print filtering image one by one according to all first feature coding values corresponding to all pixel points in each template palm print filtering image;

s2a.3.3: obtaining the target template palm print image set according to all the first feature coding value subsets corresponding to all the template palm print filtering images;

the S2b.3 specifically comprises the following steps:

s2b.3.1: in the palm print filtering image, taking the direction corresponding to the maximum value in all the second pixel accumulated values of each pixel point as a second characteristic coding value of the corresponding pixel point;

s2b.3.2: obtaining a second feature code value subset corresponding to the palm print filtering image according to all second feature code values corresponding to all pixel points in the palm print filtering image;

s2b.3.3: and obtaining the target palm print image according to the second characteristic coding value subset.

Through the coding method in the steps, the corresponding characteristic matching strategy is designed conveniently according to the target template palm print image set and the target palm print image obtained after coding, so that the subsequent characteristic extraction is conveniently carried out on the target template palm print image set and the target palm print image according to the characteristic matching strategy, the characteristic matching is carried out according to the characteristic matching strategy, the accuracy and precision of palm print identification and matching are effectively improved, and the identification and matching effect is good.

Specifically, the second intermediate template palm print ROI image is I in the embodiment_NThe (x, y) corresponding template palm print filtering image in six directions is I_Mi(x, y), on any pixel point, selecting the maximum value in the first pixel accumulated values in 6 directions, wherein the corresponding direction is the first characteristic coding value of the pixel point, traversing each pixel point to obtain the second middle template palm print ROI image as I_N(x, y) corresponding first feature code value subset, and obtaining a target template palm print image I according to the first feature code value subset_M(x, y); obtaining a target template palm print image corresponding to each second intermediate template palm print ROI image and a target palm print image J corresponding to the second intermediate palm print ROI image by adopting the same method_M(x ', y'), all of the target template palm print images constitute a set of target template palm print images.

Preferably, as shown in fig. 4, in S3, the specific step of extracting a template feature point set corresponding to each target template palm print image in the target template palm print image set includes:

s3a.1: respectively expanding each target template palm print image by adopting a bilinear interpolation method to obtain an expanded target template palm print image set;

s3a.2: in each enlarged target template palm print image of the enlarged target template palm print image set, adopting a scale invariant feature transformation method to construct a first scale space, and detecting a first extreme point set corresponding to each enlarged target template palm print image in the first scale space one by one according to a preset first pixel threshold;

s3a.3: filtering a first extreme point set of each enlarged target template palm print image in the first scale space by adopting a Harris Commer detector to obtain a template feature point set corresponding to each target template palm print image one to one;

as shown in fig. 5, in S3, the specific step of extracting the target SIFT feature point set of the target palm print image includes:

s3b.1: expanding the target palm print image by adopting a bilinear interpolation method to obtain an expanded target palm print image;

s3b.2: constructing a second scale space in the enlarged target palm print image set by adopting a scale-invariant feature transformation method, and detecting a corresponding second extreme point set of the enlarged target palm print image in the second scale space according to a preset second pixel threshold;

s3b.3: and filtering all second extreme points in the second extreme point set by adopting a Harris comber detector to obtain the target SIFT feature point set corresponding to the target palm print image.

Expanding each target template palm print image and each target palm print image by a bilinear interpolation method, so that a first extreme point set and a second extreme point set corresponding to each target template palm print image and each target palm print image can be conveniently detected by a Scale Invariant Feature Transform (SIFT) method in the follow-up process; the Harris Commer detection (Harris corner detection) method is a signal-based point feature extraction method, so that the first extreme point set and the second extreme point set are respectively filtered by the Harris Commer detector, unstable feature points can be filtered, and a template feature point set and a target SIFT feature point set with high stability are obtained; the preset pixel threshold value can be determined and adjusted according to actual conditions, and a specific operation step of filtering is performed by using a Harris Commer detector, which is a mature technology in the prior art and is not repeated herein.

It should be noted that, through the filtering of the Harris comber detector, a template feature point set corresponding to each enlarged target template palm print image one to one is obtained first, but because each enlarged target template palm print image is obtained by enlarging the corresponding target template palm print image, the template feature point set corresponding to each enlarged target template palm print image one to one is also the template feature point set corresponding to each target template palm print image one to one; similarly, the target SIFT feature point set corresponding to the target palm print image is enlarged, that is, the target SIFT feature point set corresponding to the target palm print image is enlarged.

Specifically, in the present embodiment, the target template palm print image I is subjected to_M(x, y) with the size of M multiplied by N, and obtaining an enlarged target template palm print image I by adopting a bilinear interpolation method_MD(x, y) the size of which is 2 Mx 2N, and similarly, respectively expanding each target template palm print image and each target palm print image to obtain expanded target template palm print images corresponding to each target template palm print image one by one, and J is the expanded target palm print image corresponding to each target palm print image_MD(x ', y'), wherein all of the enlarged target template palm print images constitute a set of enlarged target template palm print images.

Specifically, in the present embodiment, the target template palm print image I is enlarged_MDIn (x, y), a scale invariant feature transformation method is adopted to construct a first scale space, wherein the first scale space comprises S scales and meets the requirement of S scalesWherein k is a scale factor coefficient, S is 3, namely 3 layers of tower-shaped images are constructed, in the DOG scale space of each layer, whether the pixel gray value of each pixel point is a first extreme point in a local space range is detected according to a preset first pixel threshold value, and all the first extreme points are detected; and filtering the unstable first extreme point by adopting a Harris Commer detector to obtain a stable template characteristic point set P corresponding to the extended target template palm print image_Ij1Wherein j is₁＝1,2,…,a₁(a₁The total number of the feature points of the template feature point set corresponding to the target template palm print image is enlarged); similarly, the obtained palm print image of each enlarged target template is relatively stable in one-to-one correspondenceA template feature point set (i.e. a relatively stable template feature point set corresponding to each target template palm print image one to one), and a relatively stable target SIFT feature point set P corresponding to the enlarged target palm print image_Jj(i.e. the target palm print image corresponds to a relatively stable target SIFT feature point set), where j is 1,2, …, b (b is the total number of feature points of the target SIFT feature point set corresponding to the enlarged target palm print image).

Specifically, in this embodiment, after obtaining the more stable template feature point set and the more stable target SIFT feature point set, a second filtering operation may be performed, and for each feature point, other feature points within 4 pixels around the feature point are filtered out, so as to obtain the most stable template feature point set and the most stable target SIFT feature point set, respectively.

Preferably, as shown in fig. 6, the specific step of S4 includes:

s4.1: acquiring a template corner point coordinate set corresponding to each target template palm print image one by one according to each target template palm print image and a corresponding template feature point set based on an image pyramid method, and acquiring a target corner point coordinate set corresponding to the target palm print image according to the target palm print image and the target SIFT feature point set;

s4.2: and respectively calculating the offset between the target palm print image and each target template palm print image according to each template corner point coordinate set and the target corner point coordinate set by adopting a BLPOC (global binary offset plus one minus one plus.

An image pyramid is constructed based on an image pyramid method, template corner coordinate sets corresponding to each target template palm print image one by one and target corner coordinate sets corresponding to the target palm print images are respectively obtained in the image pyramid, and palm print deformation can be effectively corrected in a local area, so that offset between the target palm print image and each target template palm print image can be conveniently calculated subsequently, especially for a low-resolution palm print identification process, the influence of palm print deformation on palm print identification is effectively overcome, and the identification precision of palm prints is further improved; the BLPOC method (Band-limited Phase-only correlation, Band-limited Phase correlation) can effectively extract the Phase characteristics of the finger joint image, and perform the identification and matching of the finger joint print through the cross power spectrum peak of the finger joint image, so that the BLPOC method can more accurately calculate the offset between each template corner point coordinate set and the target corner point coordinate set, that is, obtain the more accurate offset between the target palm print image and each target template palm print image, thereby facilitating the subsequent identification and matching of the palm print according to the offset.

Specifically, in this embodiment, based on the image pyramid method, according to the image pyramid calculation principle, a double-layer image pyramid is constructed, and for a target template palm print image, the target template palm print image is first sampled to obtain a target template palm print image with 1/2 scales, a palm print roi (region of interest) segmentation method (for example, Otsu threshold segmentation algorithm) is adopted to segment the target template palm print image with 1/2 scales, the sampling points of each segmentation block are determined according to the degree of variation, because in general conditions, the sampling points in the palm print image are assumed to be uniformly distributed, however, the actual distribution of palm print lines is not fully considered in the assumption, the palm print texture is a more complex random structure, and the uniformly distributed sampling points can not accurately acquire the information of the palm print and can also cause the formation of redundant acquisition points; therefore, the sampling point of each partition block is selected through the degree of variation, the specific position of the sampling point is determined according to the degree of variation, and the palm print identification precision can be improved; determining the coordinates of the corner points in each partition block to obtain the corresponding coordinates of the corner points as T_It1，t₁＝1,2,…,c₁Wherein c is₁Representing the number of corner points, in terms of corner points T_I1As a reference point, the offset between the characteristic points corresponding to the target palm print image corresponding to the palm print to be matched can be calculated as O according to the double-layer image pyramid algorithm₁(ii) a Traversing each corner point, calculating the corresponding offset and marking as O_f，t₁＝1,2,…,c₁(ii) a Similarly, calculating the offset between each target template palm print image and the characteristic point corresponding to the target palm print image corresponding to the palm print to be matched, namely the offset between each original template palm print image and the original palm print image; whereinThe specific operation steps of constructing an image pyramid, segmenting an image, selecting a sampling point according to the degree of variance, acquiring a template corner coordinate set and acquiring a target corner coordinate set are all the prior art, the specific operation steps of the BLPOC method are also the prior art, and the specific details are not repeated herein.

Preferably, the specific implementation of S5 is:

When the offset is minimum, the target template palm print image corresponding to the minimum offset is closest to the target palm print image corresponding to the palm print to be matched, namely the target template palm print image is the palm print image which is most matched with the target palm print image corresponding to the palm print to be matched, and the target template palm print image is output, so that the recognition result with the highest recognition precision can be obtained, and the recognition effect is good.

Specifically, since the offset of the feature point corresponding to each target template image and the target palm print image corresponding to the palm print to be matched includes a plurality of offsets, an average operation (for example, averaging or weighted averaging) may be performed on all offsets corresponding to any target template palm print image and the target palm print image corresponding to the palm print to be matched, an obtained average result is used as a mean offset between the corresponding target template palm print image and the target palm print image, and then matching is performed according to the mean offset, specifically, the corresponding target template palm print image with the smallest mean offset is used as an identification result of the target palm print image (i.e., an identification result of the original palm print image of the palm print to be matched) and output; the specific operation method of averaging and outputting the recognition result is the existing mature technology, and is not described herein again.

In a second embodiment, as shown in fig. 7, a palm print recognition system based on an image pyramid includes an obtaining module, a preprocessing module, an extracting module, a calculating module, and a recognition module:

The method comprises the steps that original palm print images of palm prints to be matched are obtained through an obtaining module, then each original template palm print image is preprocessed through a preprocessing module respectively to obtain target template palm print images which correspond to one another, the target template palm print images form a target template palm print image set, the original palm print images are preprocessed in the same way to obtain target palm print images, through preprocessing, palm print line features affected by noise can be filtered, the noise resistance of a follow-up scale invariant feature transformation method is improved, the influence of the noise on palm print recognition is effectively overcome, meanwhile, the contrast of texture features of each target template palm print image and the target palm print images is enhanced, and follow-up palm print recognition and matching are facilitated; the target SIFT feature point set extracted by the SIFT method through the extraction module and the template feature point set corresponding to each target template palm print image one by one are further beneficial to subsequent palm print identification and matching, the influence of palm print translation on palm print identification is overcome, and the identification precision of the palm print is obviously improved; after extracting the template feature point sets corresponding to the target SIFT feature point sets and each target template palm print image one by one, the image pyramid method is adopted by the calculation module, palm print deformation can be effectively corrected in a local area, especially for the palm print recognition process with low resolution, the influence of the palm print deformation on palm print recognition is effectively overcome, the method is flexible and simple, the offset between the target palm print image and each target template palm print image can be favorably calculated subsequently, and the recognition precision of the palm print is further improved; and finally, the recognition module can judge which target template palm print image in the target template palm print image set is most matched with the target palm print image through the offset, so that the palm print to be matched is recognized, the recognition matching method is simple and effective, and the recognition precision of the palm print is obviously improved.

Preferably, as shown in fig. 8, the preprocessing module includes a region of interest extracting unit, a filtering unit and an encoding unit;

The interest extraction unit firstly extracts the interest region of each original template palm print image and the interest region of the original palm print image in the original template palm print image set, so that the effective palm print features in each palm print image can be extracted, the subsequent feature extraction and palm print identification matching are facilitated, and the operation amount can be greatly reduced; the filtering unit can effectively avoid the influence of noise on palm print identification, and the coding unit respectively codes each pixel point in each template palm print filtering image and each pixel point in the palm print filtering image, so that the subsequent characteristic extraction and the execution of the image pyramid method are facilitated, and the palm print identification and the palm print matching are facilitated.

Third, based on the first and second embodiments, the third embodiment further discloses a palm print recognition device based on the image pyramid, which includes a processor, a memory, and a computer program stored in the memory and capable of running on the processor, and the computer program realizes the specific steps S1 to S5 shown in fig. 1 when running.

The palm print recognition method is realized by the computer program stored in the memory and running on the processor, and the palm print deformation is effectively corrected based on the scale invariant feature transformation method and the image pyramid, so that the influence of the palm print deformation, the palm print translation and noise on the palm print recognition is overcome, the recognition precision of the palm print is obviously improved, and the recognition effect is good.

The present embodiment also provides a computer storage medium having at least one instruction stored thereon, where the instruction when executed implements the specific steps of S1-S5.

The palm print recognition method is realized by executing the computer storage medium containing at least one instruction, and the palm print deformation is effectively corrected based on the scale invariant feature transformation method and the image pyramid, so that the influence of the palm print deformation, palm print translation and noise on the palm print recognition is overcome, the palm print recognition precision is obviously improved, and the recognition effect is good.

Details of S1 to S5 in this embodiment are not described in detail in the first embodiment and fig. 1 to 6, which are not repeated herein.

The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.

Claims

1. A palm print identification method based on an image pyramid is characterized by comprising the following steps:

step 1: acquiring an original palm print image of a palm print to be matched;

2. The image pyramid-based palm print recognition method of claim 1, wherein in the step 2, the specific step of obtaining the target template palm print image set comprises:

3. The image pyramid-based palm print recognition method of claim 2, wherein the step 2a.2 specifically comprises:

step 2 a.2.3: in each second intermediate template palm print image of the second intermediate template palm print image set, constructing a first MFRAT filter function, establishing a p × p first filter grid by taking any pixel point as a first central point, calculating to obtain a plurality of first response values of the first central point in each direction according to the first MFRAT filter function in the first filter grid, and obtaining a first pixel accumulated value of the first central point in each direction according to all the first response values in each direction;

wherein (x, y) isCoordinates of a first central point in the two middle template palm print images, r (x, y) is a pixel value corresponding to the first central point (x, y), and theta_k(k-1, 2, …,6) is selected from six directions of 0, pi/6, 2 pi/6, 3 pi/6, 4 pi/6 and 5 pi/6,for the first filter grid at theta_kAn equation of a straight line in the direction,is the first center point (x, y) at theta_kEquation of straight line in directionA first response value of;

the step 2b.2 specifically comprises:

step 2 b.2.3: in the second intermediate palm print image, constructing a second MFRAT filtering function, establishing a p × p second filtering grid by taking any pixel point as a second central point, calculating to obtain a plurality of second response values of the second central point in each direction according to the second MFRAT filtering function in the second filtering grid, and obtaining a second pixel accumulated value of the second central point in each direction according to all the second response values in each direction;

wherein, (x ', y ') is the coordinate of the second central point in the second intermediate palm print image, r ' (x ', y ') is the pixel value corresponding to the second central point (x ', y '),for the second filter grid at θ_kAn equation of a straight line in the direction,is the second center point (x ', y') is at theta_kEquation of straight line in directionA second response value of;

4. The image pyramid-based palm print recognition method of claim 3, wherein the step 2a.3 specifically comprises:

the step 2b.3 specifically comprises:

5. The image pyramid-based palm print recognition method of claim 4, wherein in the step 3, the specific step of extracting the template feature point set corresponding to each target template palm print image in the target template palm print image set one to one comprises:

6. The image pyramid-based palm print recognition method of claim 5, wherein the specific steps of the step 4 include:

7. The image pyramid-based palm print recognition method of claim 6, wherein the step 5 is implemented by:

8. The utility model provides a palm print identification system based on image pyramid which characterized in that, includes acquisition module, preprocessing module, draws module, calculation module and identification module:

9. An image pyramid-based palm print recognition apparatus comprising a processor, a memory and a computer program stored in the memory and executable on the processor, the computer program when executed implementing the method steps according to any one of claims 1 to 7.

10. A computer storage medium, the computer storage medium comprising: at least one instruction which, when executed, implements the method steps of any one of claims 1 to 7.