Palmprint recognition method and apparatus

Abstract

Disclosed in the present invention are a palmprint recognition method and apparatus, which relate to the technical field of palmprint recognition authentication, and solve the problem of low palmprint recognition accuracy caused by incapability of correcting palm postures in the prior art. The palmprint recognition method comprises: acquiring a palm image to be recognized; performing key point detection on the palm image on the basis of a preset first key point detection method, so as to obtain a first palm key point set; constructing an affine transformation matrix on the basis of the first palm key point set; on the basis of the first palm key point set and the affine transformation matrix, performing alignment and extraction on the palm image by means of a preset alignment and extraction method, so as to obtain first palmprint information, the first palmprint information comprising a palmprint attention region and a palm key point set; and inputting the first palmprint information and pre-stored second palmprint information into a preset first recognition model, and outputting a recognition result. The present invention can align palm images and correct palm postures, and can refine the granularity of palmprint features, thus improving the precision and accuracy of palmprint recognition.

Description

Palmprint recognition method and device Technical Field

[0001] The present invention relates to the technical field of palmprint recognition and authentication, and in particular to a palmprint recognition method and device. Background Art

[0002] With the rapid development of digital technology, information security and personal authentication are becoming increasingly important. This is particularly true in areas such as financial payment, access control, attendance, and identity verification, where the demand for highly secure and reliable personal authentication methods continues to grow. Traditional authentication methods, such as passwords, PIN codes, and personal ID photos, often rely on physical or non-physical identification, which carries the risk of being forgotten, lost, or copied. Against this backdrop, biometric recognition technology is gaining attention as a natural, convenient, and difficult-to-forge authentication method.

[0003] Among the existing biometric recognition technologies, fingerprint recognition and face recognition are currently the most commonly used. Among them, fingerprint recognition has the following problems: (1) the fingerprint area is small and the amount of information is small; (2) the fingerprint is easily worn and contaminated; (3) the fingerprint is easily deformed during the collection process; and face recognition has the following problems: (1) it is greatly affected by factors such as lighting environment, obstructions, age, and facial posture; (2) there is a risk of privacy leakage. Palm prints, as one of the biometric features, have the uniqueness necessary for feature recognition. At the same time, the palm area is large and the amount of information is large. The palm print itself is stable and difficult to copy, so it can provide higher security. Existing palm print recognition technologies include structural recognition, statistical recognition, and deep learning recognition. Although these methods can achieve basic palm print recognition, the steps are relatively cumbersome and cannot correct the palm posture, resulting in low accuracy of palm print recognition.

[0004] Summary of the Invention

[0005] The present invention provides a palmprint recognition method and device to solve the technical problem in the prior art that the palmprint recognition accuracy is low due to the inability to correct the palm posture.

[0006] In a first aspect, the present application provides a palmprint recognition method, comprising:

[0007] Obtaining a palm image to be recognized;

[0008] Performing key point detection on the palm image according to a preset first key point detection method to obtain a first palm key point set;

[0009] Constructing an affine transformation matrix according to the first palm key point set;

[0010] According to the first palm key point set and the affine transformation matrix, based on a preset alignment and extraction method, the palm image is aligned and extracted to obtain first palmprint information; wherein the first palmprint information includes a palmprint attention area and a palm key point set;

[0011] The first palmprint information and the pre-stored second palmprint information are input into a preset first recognition model, and a recognition result is output.

[0012] In this way, the palm image to be identified is first obtained, and then key point detection is performed to obtain a first palm key point set, and an affine transformation matrix is ​​constructed based on the first palm key point set. Subsequently, alignment and extraction are performed based on the affine transformation matrix to obtain first palmprint information including a palmprint attention area and a palm key point set. This can align the palm image and correct the palm posture, refine the granularity of the palmprint features, and obtain and output the recognition result based on the first palmprint information and the pre-stored second palmprint information based on the first recognition model, which can improve the precision and accuracy of palmprint recognition.

[0013] Furthermore, before obtaining the palm image to be identified, the method further includes:

[0014] Get the initial palm image;

[0015] Based on a preset palm detection model, the initial palm image is detected and cropped to obtain a palm image to be identified.

[0016] In this way, the initial palm image is first obtained and then the initial palm image is detected and cropped, which can increase the information content of the palm image, reduce the influence of irrelevant areas, and reduce the error probability.

[0017] Furthermore, constructing an affine transformation matrix according to the first palm key point set specifically includes:

[0018] Based on a preset first selection method, the first palm key point set is filtered to obtain a second palm key point set;

[0019] Calculating the coordinate mean of all key points in the second palm key point set to obtain a first center point;

[0020] An affine transformation matrix is ​​constructed according to the second palm key point set and the first center point based on a preset correspondence relationship.

[0021] In this way, the key points are first filtered to obtain the second palm key point set, and the coordinates of all key points in the second palm key point set are calculated to obtain the first center point, which is convenient for the subsequent setting of the corresponding relationship and the solution of the affine transformation matrix. Then, the affine transformation matrix is ​​constructed according to the preset corresponding relationship, which can make the obtained affine transformation matrix more in line with the current alignment and extraction task scenario, and is conducive to aligning the palm image and correcting the palm posture during the subsequent alignment extraction.

[0022] Furthermore, the step of aligning and extracting the palm image based on the first palm key point set and the affine transformation matrix based on a preset alignment and extraction method to obtain first palmprint information specifically includes:

[0023] According to the second palm key point set and the first center point, based on the affine transformation matrix, the palm image is aligned and extracted to obtain a palmprint attention area;

[0024] Based on a preset second selection method, the first palm key point set is filtered to obtain a third palm key point set;

[0025] Obtaining a palm key point set according to the third palm key point set and the affine transformation matrix based on a preset key point transformation method;

[0026] The palmprint attention area is combined with the palm key point set to obtain first palmprint information.

[0027] In this way, the palm image is first aligned based on the affine transformation matrix and the palmprint attention area is extracted, which can correct the palm posture, refine the granularity of the palmprint features, obtain a more recognizable palmprint attention area, and thus reduce the probability of recognition errors. The key points are then filtered to obtain the third palm key point set, and the palm key point set is obtained based on the key point transformation method. The key point set corresponding to the aligned palm image can be obtained. Finally, the palmprint attention area and the palm key point set are combined into the first palmprint information, which can obtain the palmprint information after the palm image is aligned and the palm posture is corrected, which is convenient for subsequent palmprint recognition.

[0028] Furthermore, obtaining the palm key point set according to the third palm key point set and the affine transformation matrix based on a preset key point transformation method specifically includes:

[0029] Based on the affine transformation matrix, calculating the affine transformation points corresponding to all key points in the third palm key point set to obtain a first palm transformation point set;

[0030] Taking the transformation points of a preset sequence in the first palm transformation point set as the translation origin, calculating the translation transformation matrix of the translation origin;

[0031] Based on the translation transformation matrix, calculating the translation transformation points corresponding to all affine transformation points in the first palm transformation point set to obtain a second palm transformation point set;

[0032] Calculating a normalization coefficient based on the abscissas and ordinates of all points in the second palm transformation point set;

[0033] Based on the normalization coefficient, normalized key points corresponding to all translation transformation points in the second palm transformation point set are calculated to obtain a palm key point set.

[0034] In this way, the first palm transformation point set is obtained according to the affine transformation matrix, which can correct the palm posture and obtain the corrected key point set. Then, the translation transformation matrix is ​​calculated and the second palm transformation point set is obtained according to the translation transformation matrix. The palm image can be aligned and the image features can be simplified. Then, the normalization coefficient is calculated and the second palm transformation point set is normalized to obtain the palm key point set, which can further simplify the image features and facilitate subsequent palmprint recognition.

[0035] Furthermore, the step of inputting the first palmprint information and the pre-stored second palmprint information into a preset first recognition model and outputting a recognition result specifically includes:

[0036] Inputting the first palmprint information into the first recognition model to obtain a first recognition feature;

[0037] Inputting the second palmprint information into the first recognition model to obtain a second recognition feature;

[0038] Calculating the cosine similarity between the first identification feature and the second identification feature to obtain feature similarity;

[0039] Based on a preset threshold and the feature similarity, a recognition result is obtained and output.

[0040] In this way, the first palmprint information and the pre-stored second palmprint information are respectively input into the first recognition model to obtain the first recognition feature and the second recognition feature, and then the feature similarity is calculated by cosine similarity, and the recognition result is output according to the feature similarity. In this way, the recognition calculation using cosine similarity can reduce the recognition error and improve the accuracy of palmprint recognition.

[0041] In a second aspect, the present application provides a palmprint recognition device, comprising a palm image acquisition module, a key point detection module, a transformation matrix construction module, a palmprint information extraction module, and a palmprint verification and recognition module;

[0042] The palm image acquisition module is used to acquire the palm image to be identified;

[0043] The key point detection module is configured to perform key point detection on the palm image according to a preset first key point detection method to obtain a first palm key point set;

[0044] The transformation matrix construction module is used to construct an affine transformation matrix according to the first palm key point set;

[0045] The palmprint information extraction module is configured to align and extract the palm image based on the first palm key point set and the affine transformation matrix based on a preset alignment extraction method to obtain first palmprint information; wherein the first palmprint information includes a palmprint attention area and a palm key point set;

[0046] The palmprint verification and recognition module is used to input the first palmprint information and the pre-stored second palmprint information into a preset first recognition model and output a recognition result.

[0047] Furthermore, the palmprint recognition device further includes a palm image preprocessing module; the palm image preprocessing module includes an initial image acquisition submodule and an initial image recognition submodule;

[0048] The initial image acquisition submodule is used to acquire an initial palm image;

[0049] The initial image recognition submodule is used to detect the initial palm image based on a preset palm detection model and crop the palm image to be recognized.

[0050] Furthermore, the transformation matrix construction module includes a first screening submodule, a center point solving submodule and a matrix solving submodule;

[0051] The first screening submodule is configured to screen the first palm key point set based on a preset first selection method to obtain a second palm key point set;

[0052] The center point solving submodule is used to calculate the coordinate mean of all key points in the second palm key point set to obtain the first center point;

[0053] The matrix solving submodule is used to construct an affine transformation matrix according to the second palm key point set and the first center point based on a preset corresponding relationship.

[0054] Furthermore, the palmprint information extraction module includes an attention area extraction submodule, a second screening submodule, a key point set extraction submodule and a palmprint information combination submodule;

[0055] The attention region extraction submodule is configured to align and extract the palm image according to the second palm key point set and the first center point based on the affine transformation matrix to obtain a palmprint attention region;

[0056] The second screening submodule is configured to screen the first palm key point set based on a preset second selection method to obtain a third palm key point set;

[0057] The key point set extraction submodule is configured to obtain a palm key point set based on the third palm key point set and the affine transformation matrix and a preset key point transformation method;

[0058] The palmprint information combination submodule is used to combine the palmprint attention area with the palm key point set to obtain first palmprint information.

[0059] Furthermore, the key point set extraction submodule includes an affine transformation unit, a translation solution unit, a translation transformation unit, a normalization solution unit, and a normalization unit;

[0060] The affine transformation unit is configured to calculate, based on the affine transformation matrix, affine transformation points corresponding to all key points in the third palm key point set to obtain a first palm transformation point set;

[0061] The translation solving unit is configured to calculate a translation transformation matrix of the translation origin by taking a transformation point of a preset sequence in the first palm transformation point set as a translation origin;

[0062] The translation transformation unit is configured to calculate the translation transformation points corresponding to all affine transformation points in the first palm transformation point set based on the translation transformation matrix to obtain a second palm transformation point set;

[0063] The normalization solving unit is used to calculate a normalization coefficient according to the horizontal coordinates and vertical coordinates of all points in the second palm transformation point set;

[0064] The normalization unit is configured to calculate the normalized key points corresponding to all translation transformation points in the second palm transformation point set based on the normalization coefficient to obtain a palm key point set.

[0065] Furthermore, the palmprint verification and identification module includes a first feature acquisition submodule, a second feature acquisition submodule, a similarity calculation submodule and a verification and identification submodule;

[0066] The first feature acquisition submodule is configured to input the first palmprint information into the first recognition model to obtain a first recognition feature;

[0067] The second feature acquisition submodule is configured to input the second palmprint information into the first recognition model to obtain a second recognition feature;

[0068] The similarity calculation submodule is used to calculate the cosine similarity between the first identification feature and the second identification feature to obtain feature similarity;

[0069] The verification and recognition submodule is used to obtain and output a recognition result based on a preset threshold and the feature similarity.

[0070] In this way, the palm image to be identified is first obtained, and then key point detection is performed to obtain a first palm key point set, and an affine transformation matrix is ​​constructed based on the first palm key point set. Subsequently, alignment and extraction are performed based on the affine transformation matrix to obtain first palmprint information including a palmprint attention area and a palm key point set. This can align the palm image and correct the palm posture, refine the granularity of the palmprint features, and obtain and output the recognition result based on the first palmprint information and the pre-stored second palmprint information based on the first recognition model, which can improve the precision and accuracy of palmprint recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] FIG1 is a flow chart of an embodiment of a palmprint recognition method provided by the present invention;

[0072] FIG2 is a module structure diagram of an embodiment of a palmprint recognition device provided by the present invention;

[0073] FIG3 is a palm image obtained after key point detection according to an embodiment of a palmprint recognition method provided by the present invention;

[0074] FIG4 is a palm image after extracting the palmprint attention area according to an embodiment of a palmprint recognition method provided by the present invention;

[0075] FIG5 is a possible model structure of a first recognition model of an embodiment of a palmprint recognition method provided by the present invention. DETAILED DESCRIPTION

[0076] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0077] In the description of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly indicate the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of such features. In the description of this application, unless otherwise clearly specified, "multiple" and "several" mean two or more.

[0078] Example 1

[0079] Referring to FIG. 1 , a palmprint recognition method provided by an embodiment of the present invention includes steps S101 to S105 , each of which is specifically as follows:

[0080] Step S101: Acquire a palm image to be recognized.

[0081] Furthermore, before obtaining the palm image to be identified, the method further includes:

[0082] Get the initial palm image;

[0083] Based on a preset palm detection model, the initial palm image is detected and cropped to obtain a palm image to be identified.

[0084] In an optional embodiment, the palm detection model is preferably a target detection model, and the specific preferred model is YOLO v3.

[0085] In an optional embodiment, the palm detection model is used to detect the initial palm image and crop the palm image to be identified, specifically:

[0086] Detecting the initial palm image using a preset palm detection model to obtain a first palm image;

[0087] The first palm image is cropped according to the preset expansion parameters to obtain a palm image to be recognized.

[0088] The area to which the first palm image belongs is recorded as hand'{x,y,w,h}, where x,y are the horizontal coordinate and vertical coordinate of the upper left corner pixel of the area to which the first palm image belongs, w,h are the width and height of the area to which the first palm image belongs, and the expansion parameter is recorded as Then calculate the horizontal expansion value and vertical expansion value Then calculate the area of ​​the palm image to be recognized: hand{x-padding x , y-padding y ,w+2*padding x ,h+2*paddingy}, its size is: (w+2*padding x )*(h+2*padding y ).

[0089] In an optional embodiment, the expansion parameter The preferred value of is 0.1.

[0090] In this way, the initial palm image is first obtained and then the initial palm image is detected and cropped, which can increase the information content of the palm image, reduce the influence of irrelevant areas, and reduce the error probability.

[0091] Step S102: performing key point detection on the palm image according to a preset first key point detection method to obtain a first palm key point set.

[0092] In an optional embodiment, please refer to FIG. 3 , which is a palm image obtained after key point detection. The first palm key point set is a combination of points 0 to 20 in FIG. 3 .

[0093] In an optional embodiment, the palm image is subjected to key point detection according to a preset first key point detection method to obtain a first palm key point set, specifically:

[0094] Key point detection is performed on the palm image according to a preset key point detection model to obtain a first palm key point set; wherein the first palm key point set includes a plurality of palm key points.

[0095] Specifically, the key point detection model is preferably a MobileNet v2 model.

[0096] Specifically, the number of the plurality of palm key points can be re-determined by those skilled in the art during implementation by adjusting the key point detection model and its model parameters. The preferred number is 21.

[0097] In an optional embodiment, after obtaining the first palm key point set, the method further includes:

[0098] Based on the key point prediction model, a palmprint feature point set is obtained by screening from the first palm key point set.

[0099] Specifically, the palmprint feature point set is a combination of point 0, point 1, point 2, point 5, point 9, point 13 and point 17 in FIG3 .

[0100] Step S103: constructing an affine transformation matrix according to the first palm key point set.

[0101] Furthermore, constructing an affine transformation matrix according to the first palm key point set specifically includes:

[0102] Based on a preset first selection method, the first palm key point set is filtered to obtain a second palm key point set;

[0103] Calculating the coordinate mean of all key points in the second palm key point set to obtain a first center point;

[0104] An affine transformation matrix is ​​constructed according to the second palm key point set and the first center point based on a preset correspondence relationship.

[0105] In an optional embodiment, the first palm key point set is screened based on a preset first selection method to obtain a second palm key point set, specifically:

[0106] Based on the key point prediction model, a second palm key point set is obtained by screening from the palmprint feature point set.

[0107] Specifically, the second palm key point set is a combination of point 0, point 1, point 5 and point 17 in FIG3 .

[0108] In an optional embodiment, the calculating the coordinate mean of all key points in the second palm key point set to obtain the first center point is specifically:

[0109] Taking the second palm key point set obtained in Figure 3 as an example, point 0 is marked as (x0, y0), point 1 is (x1, y1), point 5 is (x5, y5), and point 17 is (x 17 ,y 17 ), then the first center point m is

[0110] In an optional embodiment, the affine transformation matrix is ​​constructed based on the second palm key point set and the first center point based on a preset correspondence relationship, specifically:

[0111] Considering that the size of the preset first region is r1*r2, the first region takes its upper left corner as the coordinate origin; according to the size of the first region, the corresponding relationship is set as follows: point 17 corresponds to (0,0) of the first region, point 5 corresponds to (r1,0) of the first region, and the first center point m corresponds to

[0112] According to the properties of plane affine transformation, the affine transformation matrix can be written as Among them, a, b, c, d, e, and f are all unknown parameters. According to the corresponding relationship, three sets of matrix equations can be obtained:

[0113] Based on the three sets of matrix equations, the set of expressions to be solved is:

[0114] Solve the above expression group to get the specific values ​​of a, b, c, d, e, and f, and then construct the affine transformation matrix.

[0115] In an optional embodiment, the size of the first area is preferably 112px*112px.

[0116] In this way, the key points are first filtered to obtain the second palm key point set, and the coordinates of all key points in the second palm key point set are calculated to obtain the first center point, which is convenient for the subsequent setting of the corresponding relationship and the solution of the affine transformation matrix. Then, the affine transformation matrix is ​​constructed according to the preset corresponding relationship, which can make the obtained affine transformation matrix more in line with the current alignment and extraction task scenario, and is conducive to aligning the palm image and correcting the palm posture during the subsequent alignment extraction.

[0117] Step S104: According to the first palm key point set and the affine transformation matrix, based on a preset alignment and extraction method, the palm image is aligned and extracted to obtain first palmprint information; wherein the first palmprint information includes a palmprint attention area and a palm key point set.

[0118] Furthermore, the step of aligning and extracting the palm image based on the first palm key point set and the affine transformation matrix based on a preset alignment and extraction method to obtain the first palmprint information specifically includes:

[0119] According to the second palm key point set and the first center point, based on the affine transformation matrix, the palm image is aligned and extracted to obtain a palmprint attention area;

[0120] Based on a preset second selection method, the first palm key point set is filtered to obtain a third palm key point set;

[0121] Obtaining a palm key point set according to the third palm key point set and the affine transformation matrix based on a preset key point transformation method;

[0122] The palmprint attention area is combined with the palm key point set to obtain first palmprint information.

[0123] In an optional embodiment, the palm image is aligned and extracted based on the second palm key point set and the first center point based on the affine transformation matrix to obtain the palmprint attention area, specifically:

[0124] Consider the coordinates of any pixel in the first region as (m,n), and the corresponding coordinates on the palm image as (m′,n′). According to the formula [mn 1] T =M×[m′ n′ 1] T , according to the affine transformation matrix, the corresponding mapping coordinates of all pixel points in the first area on the palm image are calculated to complete the alignment of the palm image, and the corresponding area on the palm image is extracted and cropped according to multiple mapping coordinates to obtain the palmprint attention area.

[0125] In an optional embodiment, the third palm key point set is the palmprint feature point set, including point 0, point 1, point 2, point 5, point 9, point 13 and point 17.

[0126] For the specific implementation effect, please refer to Figure 4, which is the palm image after extracting the palmprint attention area. The red box is the palmprint attention area to be extracted, the green polygon composed of points 0, 1, 5 and 17 of the second palm key point set represents the palmprint coverage area, and the red dot is the mapping point corresponding to the first center point on the palm image.

[0127] Furthermore, obtaining the palm key point set according to the third palm key point set and the affine transformation matrix based on a preset key point transformation method specifically includes:

[0128] Based on the affine transformation matrix, calculating the affine transformation points corresponding to all key points in the third palm key point set to obtain a first palm transformation point set;

[0129] Taking the transformation points of a preset sequence in the first palm transformation point set as the translation origin, calculating the translation transformation matrix of the translation origin;

[0130] Based on the translation transformation matrix, calculating the translation transformation points corresponding to all affine transformation points in the first palm transformation point set to obtain a second palm transformation point set;

[0131] Calculating a normalization coefficient based on the abscissas and ordinates of all points in the second palm transformation point set;

[0132] Based on the normalization coefficient, normalized key points corresponding to all translation transformation points in the second palm transformation point set are calculated to obtain a palm key point set.

[0133] In an optional embodiment, the affine transformation points corresponding to all key points in the third palm key point set are calculated based on the affine transformation matrix to obtain the first palm transformation point set, specifically:

[0134] The third palm key point set is recorded as {(x i ,yi )|i=0,1,2,5,9,13,17}, and the corresponding first palm transformation point set is recorded as Then the formula Calculate and obtain the first palm transformation point set.

[0135] In an optional embodiment, the calculation of the translation transformation matrix of the translation origin is performed by taking the transformation points of the preset sequence in the first palm transformation point set as the translation origin. Specifically,

[0136] The first affine transformation point in the first palm transformation point set, here point 0, is set as the translation origin (0,0). Obtain the translation transformation matrix P.

[0137] In an optional embodiment, the translation transformation points corresponding to all affine transformation points in the first palm transformation point set are calculated based on the translation transformation matrix to obtain the second palm transformation point set, specifically:

[0138] The second palm transformation point set is recorded as Then the formula Calculate and obtain the second palm transformation point set.

[0139] In an optional embodiment, the normalization coefficient is calculated according to the abscissas and ordinates of all points in the second palm transformation point set, specifically:

[0140] Selecting the horizontal coordinates and vertical coordinates of all points in the second palm transformation point set and combining them into a coordinate value set;

[0141] The reciprocal of the maximum value of the elements in the coordinate value set is used as the normalization coefficient.

[0142] In an optional embodiment, the normalized key points corresponding to all translation transformation points in the second palm transformation point set are calculated based on the normalization coefficient to obtain the palm key point set, specifically:

[0143] The horizontal coordinates and vertical coordinates of all translation transformation points in the second palm transformation point set are multiplied by a normalization coefficient to obtain a palm key point set.

[0144] In this way, the first palm transformation point set is obtained according to the affine transformation matrix, which can correct the palm posture and obtain the corrected key point set. Then, the translation transformation matrix is ​​calculated and the second palm transformation point set is obtained according to the translation transformation matrix. The palm image can be aligned and the image features can be simplified. Then, the normalization coefficient is calculated and the second palm transformation point set is normalized to obtain the palm key point set, which can further simplify the image features and facilitate subsequent palmprint recognition.

[0145] By first aligning the palm image based on the affine transformation matrix and extracting the palmprint attention area, the palm posture can be corrected, the granularity of the palmprint features can be refined, and a more recognizable palmprint attention area can be obtained, thereby reducing the probability of recognition errors. The key points are then filtered to obtain a third palm key point set, and the palm key point set is obtained based on the key point transformation method. The key point set corresponding to the aligned palm image can be obtained. Finally, the palmprint attention area and the palm key point set are combined into the first palmprint information. The palmprint information after the palm image is aligned and the palm posture is corrected can be obtained, which is convenient for subsequent palmprint recognition.

[0146] Step S105: input the first palmprint information and the pre-stored second palmprint information into a preset first recognition model, and output a recognition result.

[0147] Furthermore, the step of inputting the first palmprint information and the pre-stored second palmprint information into a preset first recognition model and outputting a recognition result specifically includes:

[0148] Inputting the first palmprint information into the first recognition model to obtain a first recognition feature;

[0149] Inputting the second palmprint information into the first recognition model to obtain a second recognition feature;

[0150] Calculating the cosine similarity between the first identification feature and the second identification feature to obtain feature similarity;

[0151] Based on a preset threshold and the feature similarity, a recognition result is obtained and output.

[0152] In an optional embodiment, the first recognition model has a preferred model structure as shown in Figure 5, wherein image_roi is the palmprint attention area, 112×112 represents the size of the palmprint attention area; landmarks_align is the palm key point set, 7*2 represents the palm key point set with a total of 7 two-dimensional plane coordinates; backboneR is a convolutional neural network, flatten is a stretching layer, backboneA is a fully connected network or a recurrent neural network, Conv2D is a two-dimensional convolution layer, concatenate is a splicing operation, backbone is a convolutional neural network or a fully connected network, and embedding is the output recognition feature.

[0153] In an optional embodiment, the preset threshold preferably ranges from 0.7 to 0.9.

[0154] In an optional embodiment, the recognition result is obtained and outputted based on the feature similarity based on a preset threshold, specifically:

[0155] When the feature similarity is not greater than the preset threshold, the recognition result is output as "verification failed";

[0156] When the feature similarity is greater than a preset threshold, the recognition result is output as "verification passed".

[0157] In this way, the first palmprint information and the pre-stored second palmprint information are respectively input into the first recognition model to obtain the first recognition feature and the second recognition feature, and then the feature similarity is calculated by cosine similarity, and the recognition result is output according to the feature similarity. In this way, the recognition calculation using cosine similarity can reduce the recognition error and improve the accuracy of palmprint recognition.

[0158] By first acquiring a palm image to be identified, then performing key point detection to obtain a first palm key point set, and constructing an affine transformation matrix based on the first palm key point set, and subsequently performing alignment and extraction based on the affine transformation matrix, first palmprint information including a palmprint attention area and a palm key point set is obtained. The palm image can be aligned and the palm posture can be corrected, the granularity of the palmprint features can be refined, and a recognition result can be obtained and output based on the first palmprint information and the pre-stored second palmprint information based on a first recognition model, thereby improving the precision and accuracy of palmprint recognition.

[0159] Example 2

[0160] 2 , which shows a palmprint recognition device according to an embodiment of the present invention, including a palm image acquisition module 201 , a key point detection module 202 , a transformation matrix construction module 203 , a palmprint information extraction module 204 , and a palmprint verification and recognition module 205 ;

[0161] The palm image acquisition module 201 is used to acquire the palm image to be identified;

[0162] The key point detection module 202 is configured to perform key point detection on the palm image according to a preset first key point detection method to obtain a first palm key point set;

[0163] The transformation matrix construction module 203 is used to construct an affine transformation matrix according to the first palm key point set;

[0164] The palmprint information extraction module 204 is configured to align and extract the palm image based on the first palm key point set and the affine transformation matrix using a preset alignment extraction method to obtain first palmprint information; wherein the first palmprint information includes a palmprint attention area and a palm key point set;

[0165] The palmprint verification and recognition module 205 is used to input the first palmprint information and the pre-stored second palmprint information into a preset first recognition model and output a recognition result.

[0166] Furthermore, the palmprint recognition device further includes a palm image preprocessing module 206; the palm image preprocessing module 206 includes an initial image acquisition submodule 261 and an initial image recognition submodule 262;

[0167] The initial image acquisition submodule 261 is used to acquire an initial palm image;

[0168] The initial image recognition submodule 262 is configured to detect the initial palm image based on a preset palm detection model and crop the palm image to be recognized.

[0169] Furthermore, the transformation matrix construction module 203 includes a first screening submodule 231, a center point solving submodule 232 and a matrix solving submodule 233;

[0170] The first screening submodule 231 is configured to screen the first palm key point set based on a preset first selection method to obtain a second palm key point set;

[0171] The center point solving submodule 232 is used to calculate the coordinate mean of all key points in the second palm key point set to obtain the first center point;

[0172] The matrix solving submodule 233 is configured to construct an affine transformation matrix based on a preset correspondence relationship according to the second palm key point set and the first center point.

[0173] Furthermore, the palmprint information extraction module 204 includes an attention region extraction submodule 241, a second screening submodule 242, a key point set extraction submodule 243 and a palmprint information combination submodule 244;

[0174] The attention region extraction submodule 241 is configured to align and extract the palm image based on the second palm key point set and the first center point based on the affine transformation matrix to obtain a palmprint attention region;

[0175] The second screening submodule 242 is configured to screen the first palm key point set based on a preset second selection method to obtain a third palm key point set;

[0176] The key point set extraction submodule 243 is configured to obtain a palm key point set based on the third palm key point set and the affine transformation matrix and a preset key point transformation method;

[0177] The palmprint information combining submodule 244 is configured to combine the palmprint attention area with the palm key point set to obtain first palmprint information.

[0178] Furthermore, the key point set extraction submodule includes an affine transformation unit, a translation solution unit, a translation transformation unit, a normalization solution unit, and a normalization unit;

[0179] The affine transformation unit is configured to calculate, based on the affine transformation matrix, affine transformation points corresponding to all key points in the third palm key point set to obtain a first palm transformation point set;

[0180] The translation solving unit is configured to calculate a translation transformation matrix of the translation origin by taking a transformation point of a preset sequence in the first palm transformation point set as a translation origin;

[0181] The translation transformation unit is configured to calculate the translation transformation points corresponding to all affine transformation points in the first palm transformation point set based on the translation transformation matrix to obtain a second palm transformation point set;

[0182] The normalization solving unit is used to calculate a normalization coefficient according to the horizontal coordinates and vertical coordinates of all points in the second palm transformation point set;

[0183] The normalization unit is configured to calculate the normalized key points corresponding to all translation transformation points in the second palm transformation point set based on the normalization coefficient to obtain a palm key point set.

[0184] Furthermore, the palmprint verification and identification module 205 includes a first feature acquisition submodule 251, a second feature acquisition submodule 252, a similarity calculation submodule 253 and a verification and identification submodule 254;

[0185] The first feature acquisition submodule 251 is used to input the first palmprint information into the first recognition model to obtain a first recognition feature;

[0186] The second feature acquisition submodule 252 is used to input the second palmprint information into the first recognition model to obtain a second recognition feature;

[0187] The similarity calculation submodule 253 is used to calculate the cosine similarity between the first identification feature and the second identification feature to obtain feature similarity;

[0188] The verification and recognition submodule 254 is configured to obtain and output a recognition result based on a preset threshold and the feature similarity.

[0189] In this way, the palm image to be identified is first obtained, and then key point detection is performed to obtain a first palm key point set, and an affine transformation matrix is ​​constructed based on the first palm key point set. Subsequently, alignment and extraction are performed based on the affine transformation matrix to obtain first palmprint information including a palmprint attention area and a palm key point set. This can align the palm image and correct the palm posture, refine the granularity of the palmprint features, and obtain and output the recognition result based on the first palmprint information and the pre-stored second palmprint information based on the first recognition model, which can improve the precision and accuracy of palmprint recognition.

[0190] Correspondingly, the embodiment of the present invention also adaptively provides a terminal device and a computer-readable storage medium.

[0191] The terminal device includes: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor;

[0192] When the processor executes the computer program, the palmprint recognition method as described above is implemented.

[0193] The computer-readable storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor to execute the palmprint recognition method as described above.

[0194] The specific embodiments described above further illustrate the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A palmprint recognition method, characterized in that: include: Obtaining a palm image to be recognized; Performing key point detection on the palm image according to a preset first key point detection method to obtain a first palm key point set; Constructing an affine transformation matrix according to the first palm key point set; According to the first palm key point set and the affine transformation matrix, based on a preset alignment and extraction method, the palm image is aligned and extracted to obtain first palmprint information; wherein the first palmprint information includes a palmprint attention area and a palm key point set; The first palmprint information and the pre-stored second palmprint information are input into a preset first recognition model, and a recognition result is output.

2. A palmprint recognition method according to claim 1, characterized in that: Before obtaining the palm image to be identified, the method further includes: Get the initial palm image; Based on a preset palm detection model, the initial palm image is detected and cropped to obtain a palm image to be identified.

3. A palmprint recognition method according to claim 1, characterized in that: The constructing of an affine transformation matrix according to the first palm key point set specifically includes: Based on a preset first selection method, the first palm key point set is filtered to obtain a second palm key point set; Calculating the coordinate mean of all key points in the second palm key point set to obtain a first center point; An affine transformation matrix is ​​constructed according to the second palm key point set and the first center point based on a preset correspondence relationship.

4. A palmprint recognition method according to claim 3, characterized in that: The step of aligning and extracting the palm image based on the first palm key point set and the affine transformation matrix to obtain first palmprint information specifically includes: According to the second palm key point set and the first center point, based on the affine transformation matrix, the palm image is aligned and extracted to obtain a palmprint attention area; Based on a preset second selection method, the first palm key point set is filtered to obtain a third palm key point set; Obtaining a palm key point set according to the third palm key point set and the affine transformation matrix based on a preset key point transformation method; The palmprint attention area is combined with the palm key point set to obtain first palmprint information.

5. A palmprint recognition method according to claim 4, characterized in that: The step of obtaining the palm key point set according to the third palm key point set and the affine transformation matrix based on a preset key point transformation method specifically includes: Based on the affine transformation matrix, calculating the affine transformation points corresponding to all key points in the third palm key point set to obtain a first palm transformation point set; Taking the transformation points of a preset sequence in the first palm transformation point set as the translation origin, calculating the translation transformation matrix of the translation origin; Based on the translation transformation matrix, calculating the translation transformation points corresponding to all affine transformation points in the first palm transformation point set to obtain a second palm transformation point set; Calculating a normalization coefficient based on the abscissas and ordinates of all points in the second palm transformation point set; Based on the normalization coefficient, normalized key points corresponding to all translation transformation points in the second palm transformation point set are calculated to obtain a palm key point set.

6. A palmprint recognition method according to claim 1, characterized in that: The step of inputting the first palmprint information and the pre-stored second palmprint information into a preset first recognition model and outputting a recognition result specifically includes: Inputting the first palmprint information into the first recognition model to obtain a first recognition feature; Inputting the second palmprint information into the first recognition model to obtain a second recognition feature; Calculating the cosine similarity between the first identification feature and the second identification feature to obtain feature similarity; Based on a preset threshold and the feature similarity, a recognition result is obtained and output.

7. A palmprint recognition device, characterized in that: It includes palm image acquisition module, key point detection module, transformation matrix construction module, palm print information extraction module and palm print verification and recognition module; The palm image acquisition module is used to acquire the palm image to be identified; The key point detection module is configured to perform key point detection on the palm image according to a preset first key point detection method to obtain a first palm key point set; The transformation matrix construction module is used to construct an affine transformation matrix according to the first palm key point set; The palmprint information extraction module is configured to align and extract the palm image based on the first palm key point set and the affine transformation matrix based on a preset alignment extraction method to obtain first palmprint information; wherein the first palmprint information includes a palmprint attention area and a palm key point set; The palmprint verification and recognition module is used to input the first palmprint information and the pre-stored second palmprint information into a preset first recognition model and output a recognition result.

8. A palmprint recognition device according to claim 7, characterized in that: The palmprint recognition device further includes a palm image preprocessing module; the palm image preprocessing module includes an initial image acquisition submodule and an initial image recognition submodule; The initial image acquisition submodule is used to acquire an initial palm image; The initial image recognition submodule is used to detect the initial palm image based on a preset palm detection model and crop the palm image to be recognized.

9. The palmprint recognition device according to claim 7, characterized in that: The transformation matrix construction module includes a first screening submodule, a center point solving submodule and a matrix solving submodule; The first screening submodule is configured to screen the first palm key point set based on a preset first selection method to obtain a second palm key point set; The center point solving submodule is used to calculate the coordinate mean of all key points in the second palm key point set to obtain the first center point; The matrix solving submodule is used to construct an affine transformation matrix according to the second palm key point set and the first center point based on a preset corresponding relationship.

10. The palmprint recognition device according to claim 9, characterized in that: The palmprint information extraction module includes an attention area extraction submodule, a second screening submodule, a key point set extraction submodule and a palmprint information combination submodule; The attention region extraction submodule is configured to align and extract the palm image according to the second palm key point set and the first center point based on the affine transformation matrix to obtain a palmprint attention region; The second screening submodule is configured to screen the first palm key point set based on a preset second selection method to obtain a third palm key point set; The key point set extraction submodule is configured to obtain a palm key point set based on the third palm key point set and the affine transformation matrix and a preset key point transformation method; The palmprint information combination submodule is used to combine the palmprint attention area with the palm key point set to obtain first palmprint information.

11. A palmprint recognition device according to claim 10, characterized in that: The key point set extraction submodule includes an affine transformation unit, a translation solution unit, a translation transformation unit, a normalization solution unit, and a normalization unit; The affine transformation unit is configured to calculate, based on the affine transformation matrix, affine transformation points corresponding to all key points in the third palm key point set to obtain a first palm transformation point set; The translation solving unit is configured to calculate a translation transformation matrix of the translation origin by taking a transformation point of a preset sequence in the first palm transformation point set as a translation origin; The translation transformation unit is configured to calculate the translation transformation points corresponding to all affine transformation points in the first palm transformation point set based on the translation transformation matrix to obtain a second palm transformation point set; The normalization solving unit is used to calculate a normalization coefficient according to the horizontal coordinates and vertical coordinates of all points in the second palm transformation point set; The normalization unit is used to calculate the corresponding values ​​of all translation transformation points in the second palm transformation point set based on the normalization coefficient. Normalize the key points to obtain the palm key point set.

12. The palmprint recognition device according to claim 7, characterized in that: The palmprint verification and identification module includes a first feature acquisition submodule, a second feature acquisition submodule, a similarity calculation submodule and a verification and identification submodule; The first feature acquisition submodule is configured to input the first palmprint information into the first recognition model to obtain a first recognition feature; The second feature acquisition submodule is configured to input the second palmprint information into the first recognition model to obtain a second recognition feature; The similarity calculation submodule is used to calculate the cosine similarity between the first identification feature and the second identification feature to obtain feature similarity; The verification and recognition submodule is used to obtain and output a recognition result based on a preset threshold and the feature similarity.

Images