WO2020253505A1

WO2020253505A1 - Palm image detection method and apparatus

Info

Publication number: WO2020253505A1
Application number: PCT/CN2020/093510
Authority: WO
Inventors: 杨祎; 王炜
Original assignee: 平安科技（深圳）有限公司
Priority date: 2019-06-20
Filing date: 2020-05-29
Publication date: 2020-12-24
Also published as: CN110414330B; CN110414330A

Abstract

The present application relates to the technical field of image processing. Disclosed are a palm image detection method and apparatus. The method comprises: using a solid mask for marking and obtaining a palm image in a training image; intercepting sub-images in the training image; determining positive sample images and negative sample images in the sub-images, counting the number of positive samples and the number of negative samples, determining whether preset conditions are satisfied, and if yes, intercepting the sub-images in the training image again; if not, extracting the positive sample images and the negative sample images, and generating a sample training set; training the sample training set by means of a Fast R-CNN model, and generating weight parameters of the training image; calculating a feature vector value of an image to be detected according to the weight parameters; and determining whether the image to be detected comprises the palm image according to the feature vector value. By means of the approach, the accuracy of palm image recognition can be improved.

Description

Method and device for detecting palm image

This application affirms that it enjoys the priority of a Chinese patent application filed on June 20, 2019 with the application number 201910534814.6 and titled "A method and device for detecting palm images". The entire content of the Chinese patent application is incorporated by reference In this application.

Technical field

This application relates to the technical field of image processing based on neural networks, in particular to a method and device for detecting palm images.

Background technique

With the rapid development of information technology, palmprint recognition technology has been widely used in various identity authentication scenarios due to its reliability and convenience. The palmprint recognition technology uses the characteristics of human palm patterns to identify the identity. The specific process includes: first take a picture of the user's palm to be tested, and then extract the palmprints of the palm of the palm to be tested and match the palmprints stored in the system. If the matching is successful, the palmprint recognition is considered successful, and the user's identity is legal.

In the process of extracting palm prints in the palm image to be tested, it is usually necessary to first determine whether the captured image contains valid palm prints, and then perform subsequent palm print extraction operations. In the prior art, the edge of the image is detected first, and then the training results of the effective palm image and the invalid palm image by the SVM classifier are used to determine whether the palm image to be tested contains a valid palm image. The SVM classifier is trained using image gray gradient information. of.

technical problem

The inventor realizes that if the palm image to be tested includes palm, arm, and face images with similar colors, especially when images with similar colors overlap, the gray information corresponding to the images with similar colors is basically the same, and gray information cannot be passed. The palm, arm and face are distinguished, so it is difficult to accurately judge whether the palm image to be tested contains a valid palm image by using the gray gradient information.

Technical solutions

In view of this, the present application provides a palm image detection method and device, the main purpose of which is to solve the problem of low accuracy in determining whether the palm image to be tested includes a palm image in the prior art.

According to one aspect of the present application, a method for detecting palm images is provided, including: using a solid mask to mark a palm image in a training image, and obtaining the palm image; within the image pixel range of the training image, according to random The image size and pixel position calculated by the function are used to intercept the sub-images in the training image; if the sub-image contains the palm image, the sub-image is determined to be a positive sample image, and statistics of the positive sample image The number of positive samples; if the sub-image does not completely contain the palm image, the sub-image is determined to be a negative sample image, and the number of negative samples of the negative sample image is counted; if the number of positive samples is less than the preset minimum The number of positive samples, or the number of negative samples is less than the preset minimum number of negative samples, then the sub-images in the training image are captured again; if the number of positive samples is not less than the preset minimum number of positive samples, and the If the number of negative samples is not less than the preset minimum number of negative samples, the positive sample image and the negative sample image are extracted according to the preset total number of positive and negative samples and the preset ratio of positive and negative samples to generate a sample training set; The region-based fast convolutional neural network Fast R-CNN model trains the positive sample image and the negative sample image in the sample training set, and generates weight parameters of image features of the training image, the image features including shapes , Color and shadow; calculate the feature vector value of the image to be tested according to the weight parameter; determine whether the palm image is included in the image to be tested according to the feature vector value.

According to another aspect of the present application, there is provided a palm image detection device, including: a marking module for marking the palm image in the training image with a solid mask, and obtaining the palm image; Within the image pixel range of the training image, the sub-image in the training image is intercepted according to the image size and pixel position calculated by a random function; the determining module is used to determine if the sub-image contains the palm image The sample image is a positive sample image, and the number of positive samples of the positive sample image is counted; the determining module is further configured to determine that the sample image is a negative sample if the sub-image does not completely contain the palm image Image, and count the number of negative samples of the negative sample image; the intercepting module is used for if the number of positive samples is less than the preset minimum number of positive samples, or the number of negative samples is less than the preset minimum number of negative samples, then Intercepting the sub-images in the training image again; an extraction module for if the number of positive samples is not less than the preset minimum number of positive samples, and the number of negative samples is not less than the preset minimum number of negative samples, According to the preset total number of positive and negative samples and the preset ratio of positive and negative samples, the positive sample image and the negative sample image are extracted to generate a sample training set; the generation module is used to pass the region-based fast convolutional neural network Fast The R-CNN model trains the positive sample image and the negative sample image in the sample training set, and generates the weight parameters of the image features of the training image. The image features include shape, color, and shadow; the calculation module uses According to the weight parameter, the feature vector value of the image to be measured is calculated; the judgment module is used for judging whether the palm image is included in the image to be measured according to the feature vector value.

According to another aspect of the present application, there is provided a storage medium in which at least one executable instruction is stored, and the executable instruction causes a processor to perform the following steps: using a solid mask to mark the training image The palm image of the training image, and obtain the palm image; within the image pixel range of the training image, according to the image size and pixel position calculated by the random function, intercept the sub-image in the training image; if the sub-image contains For the palm image, determine that the sub-image is a positive sample image, and count the number of positive samples of the positive sample image; if the sub-image does not completely contain the palm image, determine that the sub-image is a negative sample Image, and count the number of negative samples of the negative sample image; if the number of positive samples is less than the preset minimum number of positive samples, or the number of negative samples is less than the preset minimum number of negative samples, then the training image is captured again If the number of positive samples is not less than the preset minimum number of positive samples, and the number of negative samples is not less than the preset minimum number of negative samples, according to the preset total number of positive and negative samples and the preset Positive-negative sample ratio, extract the positive sample image and the negative sample image to generate a sample training set; train the positive sample image and the sample training set in the sample training set through the region-based fast convolutional neural network Fast R-CNN model The negative sample image generates the weight parameters of the image features of the training image, the image features include shape, color, and shadow; according to the weight parameters, the feature vector value of the image to be tested is calculated; according to the feature vector value To determine whether the image to be tested contains the palm image.

According to another aspect of the present application, there is provided a computer device, including: a processor, a memory, a communication interface, and a communication bus. The processor, the memory, and the communication interface complete mutual communication through the communication bus. Communication; the memory is used to store at least one executable instruction, the executable instruction causes the processor to perform the following steps: use a solid mask to mark the palm image in the training image, and obtain the palm image; Within the image pixel range of the training image, the sub-image in the training image is intercepted according to the image size and pixel position calculated by a random function; if the sub-image includes the palm image, it is determined that the sub-image is Positive sample image, and count the number of positive samples of the positive sample image; if the sub-image does not completely contain the palm image, determine that the sub-image is a negative sample image, and count the negative samples of the negative sample image Number; if the number of positive samples is less than the preset minimum number of positive samples, or the number of negative samples is less than the preset minimum number of negative samples, then the sub-images in the training image are captured again; if the number of positive samples is not less than If the preset minimum number of positive samples and the number of negative samples are not less than the preset minimum number of negative samples, the positive sample image and the preset positive and negative sample ratio are extracted according to the preset total number of positive and negative samples. The negative sample image generates a sample training set; the positive sample image and the negative sample image in the sample training set are trained by the region-based fast convolutional neural network Fast R-CNN model to generate the training image The weight parameter of the image feature, the image feature includes shape, color and shadow; according to the weight parameter, the feature vector value of the image to be measured is calculated; according to the feature vector value, it is judged whether the image to be measured contains the palm image .

Beneficial effect

The technical solution provided by the embodiments of the present application obtains positive sample images or negative sample images by marking the palm images in the training images to improve the accuracy of the training samples, so as to improve the accuracy of the weight parameters of image features obtained by training based on the training samples. By training the weight parameters of image features such as shape, color, and shadow in the Fast R-CNN model, the accuracy of judging whether the image to be tested contains palm images according to the weight parameters is improved.

Description of the drawings

By reading the detailed description of the preferred embodiments below, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only used for the purpose of illustrating the preferred embodiments, and are not considered as a limitation to the application. Also, throughout the drawings, the same reference symbols are used to denote the same components. In the attached picture:

Fig. 1 shows a flowchart of a palm image detection method provided by an embodiment of the present application;

FIG. 2 shows a flowchart of another palm image detection method according to an embodiment of the present application;

Figure 3 shows a block diagram of a palm image detection device provided by an embodiment of the present application;

FIG. 4 shows a block diagram of another palm image detection composition provided by an embodiment of the present application;

Fig. 5 shows a schematic structural diagram of a computer device provided by an embodiment of the present application.

Embodiments of the invention

Hereinafter, exemplary embodiments of the present disclosure will be described in more detail with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited by the embodiments set forth herein.

In the palmprint recognition process, it is usually necessary to first determine whether the captured image contains a valid palmprint pattern. If it is determined that the captured image contains a valid palmprint pattern, then perform subsequent palmprint extraction and recognition operations. In the actual image to be measured, in addition to the palm, it may also include images of objects that are similar in color to the palm, and the images of near-colored objects may also overlap, so that the hands in the image to be measured can be accurately identified and segmented. The shape is particularly difficult. The purpose of the embodiments of the present application is to improve the accuracy of determining whether a palm image is included in the image to be tested. The embodiment of the application provides a method for detecting a palm image. As shown in FIG. 1, the method includes:

101. Use a solid mask to mark the palm image in the training image, and obtain the palm image.

The training image is a preset image that can fully and correctly identify the palm image in the image. The palm image is not a specific image with the same palmprint features, but an image corresponding to a real palm in any image. In the embodiment of the present application, it refers to the palm image in the training image or the palm image in the image to be tested. Masking in image processing refers to the use of selected images, graphics or objects to block the training image in order to control the area or process of image processing. The solid mask means that when the training image is occluded, the area within the marked pixels is completely occluded, and the occluded area is the palm image.

Use existing image processing tools to mark the palm image in the training image, and multiply the palm image mask with the training image to obtain the palm image in the training image. The pixel value in the palm image remains unchanged, while the image value outside the palm image Both are 0. That is, the mask of the palm image masks the area outside the palm image. An exemplary description of the method of extracting a palm image from a training image is to perform a dot multiplication operation on each pixel in the training image and each corresponding pixel of the palm image mask. For example, the pixel matrix of the training image is [23,22 ,89;0,0,255;90,0,23], the palm image mask is [0,0,1;1,0,1;1,1,1], the pixel matrix of the palm image obtained after the dot product is [0,0,89;0,0,255;90,0,23].

102. Within the image pixel range of the training image, intercept sub-images in the training image according to the image size and pixel position calculated by a random function.

The sub-image is a part of the training image, and the size of the sub-image is not limited in the embodiment of the present application. Use a random function to calculate the first random value within the image pixel range of the training image, the first random value is the image size of the sub-image, the shape of the sub-image is rectangular, and the first random value includes the width and height of the sub-image . A random function is used to calculate a second random value within the image pixel range of the training image, and the second random value is the pixel position of the sub-image. Then take the pixel position as the starting point and the image size as the selected area to intercept the sub-images in the training image. When capturing an image, first set the starting point of which vertex of the selected area, the vertex of the selected area can be the upper left corner, the lower left corner, the upper right corner or the lower right corner, and then determine the set vertex position and the size of the selected area , To intercept the sub-images in the training image.

103. If the sub-image includes the palm image, determine that the sub-image is a positive sample image, and count the number of positive samples of the positive sample image.

The sub-image is compared with the palm image to determine whether the sub-image contains the palm image. If the palm image is completely contained in the sub-image, the sub-image is determined to be a positive sample image. In the process of judging whether the sub-image contains the palm image, you can first obtain the mark position of the palm image in the training image, and then calculate the image boundary of the sub-image in the training image according to the fixed point and selected area size of the sub-image in the training image , According to the image boundary, it is judged whether the label positions all fall within the position range of the sub-image. If the judgment result is yes, the sub-image is determined to be a positive sample image. If the sub-image is a positive sample image, the value of the number of positive samples corresponding to the positive sample image is increased by one.

104. If the sub-image does not completely include the palm image, determine that the sub-image is a negative sample image, and count the number of negative samples of the negative sample image.

This step is parallel to step 103 and is similar to step 103. The process of judging whether the sub-image contains the palm image is the same. If the sub-image does not completely contain the palm image, that is, the marked position of the palm image in the training image is not all If it falls within the image boundary corresponding to the sub-image, it is determined that the sub-image is a negative sample image. If the sub-image is a negative sample image, the value of the number of negative samples corresponding to the negative sample image is increased by one.

105. If the number of positive samples is less than the preset minimum number of positive samples, or the number of negative samples is less than the preset minimum number of negative samples, then the sub-images in the training image are captured again.

Every time a sub-image is taken, after determining whether the sub-image is a positive sample image or a negative sample image, it is determined whether the number of positive samples is less than the preset minimum number of samples, and whether the number of negative samples is less than the preset minimum number of negative samples. If the number of positive samples is less than the preset minimum number of positive samples, or the number of negative samples is less than the preset minimum number of negative samples, it means that the number of positive sample images and negative sample images does not meet the training requirements, and you need to intercept the training images again according to step 102 Sub-image.

106. If the number of positive samples is not less than the preset minimum number of positive samples, and the number of negative samples is not less than the preset minimum number of negative samples, follow the preset total number of positive and negative samples and the preset number of positive and negative samples. Sample ratio, extract the positive sample image and the negative sample image, and generate a sample training set.

This step is similar to the method described in step 105. Every time a sub-image is captured, after judging whether the sub-image is a positive sample image or a negative sample image, it is determined whether the number of positive samples is less than the preset minimum number of samples, and whether the number of negative samples is less than Preset the minimum number of negative samples. If the number of positive samples is not less than the preset minimum number of positive samples, and the number of negative samples is not less than the preset minimum number of negative samples, that is, when both the number of positive samples and the number of negative samples meet the training requirements, the operation of this step is executed. The sample training set includes positive sample images and negative sample images, and the sample training set needs to include a sufficient number of positive sample images and negative sample images. The preset total number of positive and negative samples refers to the sum of all positive and negative sample images in the sample training set, and the preset positive and negative sample ratio refers to the ratio of the number of positive samples to the number of negative samples in the sample training set. According to the preset total number of positive and negative samples and the preset ratio of positive and negative samples, calculate the required number of positive samples and the required number of negative samples in the sample training set, and extract the positive sample image of the required number of positive samples and the negative sample image of the required number of negative samples, Generate sample training set.

107. Train the positive sample image and the negative sample image in the sample training set by using a region-based fast convolutional neural network Fast R-CNN model to generate weight parameters of image features of the training image.

Input the positive sample images and negative sample images in the sample training set into the region-based fast convolutional neural network Fast R-CNN model, and continuously adjust and generate training images for the purpose of detecting the palm images in the positive sample images and negative sample images The weight parameter of the image features, where the image features include shape, color, and shadow. The special shape of the palm, the color that is basically the same as that of the human face and arm, and the shadow caused by the different light source angles are important image features that affect the detection of the palm image. By training positive sample images and negative sample images that are known to be palm images, and constantly adjusting the weight parameters corresponding to the image features, the Fast R-CNN model has a higher recognition accuracy.

108. Calculate the feature vector value of the image to be tested according to the weight parameter.

The image to be tested refers to an image that requires palmprint recognition, and is usually an image taken immediately before palmprint recognition. According to a preset algorithm, the image features of the image to be tested are calculated. The image features include shape, color and shadow, and then the feature vector value of the image to be tested is calculated according to the weight parameter.

109. According to the feature vector value, determine whether the image to be tested includes the palm image.

When the feature vector value is greater than the preset threshold, the image to be tested includes a palm image, and when the feature vector value is not greater than the preset threshold, the image to be tested does not include the palm image.

Determine whether the image to be tested contains a palm image, and then perform the next operation according to the result of the judgment. If the result of the judgment is yes, identify whether the palm image in the image to be tested is a palm image that has been recorded. Measure the image. Of course, if the judgment result is no, it can also prompt that no palm is detected.

In the embodiment of the present application, a positive sample image or a negative sample image is obtained by marking the palm image in the training image to improve the accuracy of the training sample, so as to improve the accuracy of the weight parameter of the image feature obtained by training based on the training sample. By training the weight parameters of image features such as shape, color, and shadow in the Fast R-CNN model, the accuracy of judging whether the image to be tested contains palm images according to the weight parameters is improved.

The embodiment of the present application provides another palm image detection method. As shown in FIG. 2, the method includes:

201. Use a solid mask to mark a palm image in a training image, and obtain the palm image.

The solid mask means that when the training image is occluded, the area within the marked pixels is completely occluded, and the occluded area is the palm image. Obtaining the palm image is to obtain the palm image in the marked area of the training image. The specific process includes: using a solid mask to mark the palm image in the training image; obtaining the pixel coordinates of the marked pixels in the training image marked by the solid mask; generating a mask according to the pixel coordinates Matrix; Do a dot product operation on the mask matrix and the training image to generate and acquire the palm image.

202. Within the image pixel range of the training image, intercept sub-images in the training image according to the image size and pixel position calculated by a random function.

This step is similar to the method described in step 102 shown in FIG. 1, and will not be repeated here.

203. If the sub-image includes the palm image, determine that the sub-image is a positive sample image, and count the number of positive samples of the positive sample image.

This step is similar to the method described in step 103 shown in FIG. 1, and will not be repeated here.

204. If the sub-image does not completely include the palm image, determine that the sub-image is a negative sample image, and count the number of negative samples of the negative sample image.

This step is parallel to step 203, and is similar to the method described in step 104 shown in FIG. 1, and will not be repeated here.

205. If the number of positive samples is less than the preset minimum number of positive samples, or the number of negative samples is less than the preset minimum number of negative samples, then the sub-images in the training image are captured again.

This step is similar to the method described in step 105 shown in FIG. 1, and will not be repeated here.

206. If the number of positive samples is not less than the preset minimum number of positive samples, and the number of negative samples is not less than the preset minimum number of negative samples, follow the preset total number of positive and negative samples and the preset number of positive and negative samples. Sample ratio, extract the positive sample image and the negative sample image, and generate a sample training set.

Every time a sub-image is taken, after determining whether the sub-image is a positive sample image or a negative sample image, it is determined whether the number of positive samples is less than the preset minimum number of samples, and whether the number of negative samples is less than the preset minimum number of negative samples. This step specifically includes: saving the positive sample image in the positive sample library, and saving the negative sample image in the negative sample library; according to the preset total number of positive and negative samples and the positive and negative sample ratio, according to the preset The rule extracts the positive sample image and the negative sample image from the positive sample library and the negative sample library respectively, and generates a sample training set.

The preset rules for extracting positive sample images from the positive sample library, or extracting negative sample images from the negative sample library, can be extracted sequentially in the order of storage, can be extracted in the order of size from small to large, or randomly Extraction, the extraction rules for extracting positive sample images or negative sample images are not limited in the embodiment of the present application. In the extraction process, the ratio of positive and negative samples used can be 1:3. The set extracts positive sample images and negative sample images from the positive sample library and the negative sample library, respectively, to generate a sample training set.

207. Train the positive sample image and the negative sample image in the sample training set by using a region-based fast convolutional neural network Fast R-CNN model to generate weight parameters of image features of the training image.

The process of training the sample training set is to obtain and continuously modify the weight parameters of the image features according to the known judgment result of whether the palm image is included in the sample training set, so that the Fast R-CNN model has a higher recognition accuracy.

This step specifically includes: in the Fast R-CNN model, input the positive sample image or the negative sample image in the training sample set; adopt the regional candidate network to calculate and generate the preset of the positive sample image or the negative sample image Suggestion window of the number of windows; map the suggestion window to the feature map convolutional layer of the Fast R-CNN model; combine the feature map convolutional layer with each of the RoI pooling layer of the Fast R-CNN model A convolution operation is performed on a rectangular ROI to generate a feature map of a fixed size; the classification probability of the feature map of a fixed size and frame regression are jointly trained to generate the weight parameter of the image feature.

Among them, during joint training: use the detection classification probability Softmax Loss and the detection frame regression Smooth L1 Loss to jointly train the fixed size feature for the classification probability and the frame regression respectively map, generating the weight parameter of the image feature.

208. Calculate the feature vector value of the image to be tested according to the weight parameter.

209. According to the feature vector value, determine whether the image to be tested includes the palm image.

Specifically: if the feature vector value is greater than a preset threshold, determining that the image to be tested contains the palm image; if the feature vector value is not greater than the preset threshold, then determining that the image to be tested is The palm image is not included. If the judgment result is otherwise, the image to be tested needs to be retaken. Of course, if the judgment result is no, the shooting camera can be automatically restarted for shooting.

210. If the palm image is included in the image to be tested, identify palmprint feature information in the palm image.

Due to the large amount of information contained in the palm image, in order to improve the speed of palmprint recognition, it is first necessary to identify the palmprint feature information in the palm image. The palmprint feature information includes main lines, folds, minutiae points and triangle points. The palm image in this step refers to the palm image included in the image to be tested. The method of recognizing palmprint feature information used in this step is the same as the method of recognizing palmprint feature information entered by the user, so as to improve the recognition accuracy.

211. Identify the image to be tested according to the palmprint feature information.

Recognizing the image to be tested is to identify whether the palmprint feature information is the same as the palmprint feature information saved by the recognition system. If they are the same, the image to be tested can be recognized to obtain the system authority of the response. If the image to be tested contains a palm image, but it is not recognized by the palmprint recognition system, it can also detect whether the continuously shot images to be tested are the same image. If yes, an alarm message is generated and the alarm message is sent. Send the alarm information to the administrator to notify the administrator or wait for the administrator to activate emergency measures such as the shooting and recognition functions of the image to be tested no longer in the preset time period, providing double guarantee for the safety of the palmprint recognition system .

Further, as an implementation of the method shown in FIG. 1, an embodiment of the present application provides a palm image detection device. As shown in FIG. 3, the device includes:

The marking module 31 is configured to use a solid mask to mark the palm image in the training image, and obtain the palm image;

The interception module 32 is configured to intercept the sub-images in the training image according to the image size and pixel position calculated by a random function within the image pixel range of the training image;

The determining module 33 is configured to determine that the sample image is a positive sample image if the sub-image includes the palm image, and count the number of positive samples of the positive sample image;

The determining module 33 is further configured to determine that the sample image is a negative sample image if the sub-image does not completely include the palm image, and count the number of negative samples of the negative sample image;

The intercepting module 32 is configured to intercept the sub-images in the training image again if the number of positive samples is less than the preset minimum number of positive samples, or the number of negative samples is less than the preset minimum number of negative samples;

The extraction module 34 is configured to, if the number of positive samples is not less than the preset minimum number of positive samples, and the number of negative samples is not less than the preset minimum number of negative samples, follow the preset total number of positive and negative samples Preset the ratio of positive and negative samples, extract the positive sample image and the negative sample image, and generate a sample training set;

The generating module 35 is configured to train the positive sample image and the negative sample image in the sample training set through the region-based fast convolutional neural network Fast R-CNN model, and generate the weight parameters of the image features of the training image , The image features include shape, color and shadow;

The calculation module 36 is configured to calculate the feature vector value of the image to be tested according to the weight parameter;

The determining module 37 is configured to determine whether the palm image is included in the image to be tested according to the feature vector value.

Further, as an implementation of the method shown in FIG. 2, an embodiment of the present application provides another palm image detection device. As shown in FIG. 4, the device includes:

The marking module 41 is configured to use a solid mask to mark the palm image in the training image, and obtain the palm image;

The intercepting module 42 is configured to intercept the sub-images in the training image according to the image size and pixel position calculated by the random function within the image pixel range of the training image;

The determining module 43 is configured to determine that the sample image is a positive sample image if the sub-image includes the palm image, and count the number of positive samples of the positive sample image;

The determining module 43 is further configured to determine that the sample image is a negative sample image if the sub-image does not completely include the palm image, and count the number of negative samples of the negative sample image;

The interception module 42 is configured to intercept the sub-images in the training image again if the number of positive samples is less than the preset minimum number of positive samples, or the number of negative samples is less than the preset minimum number of negative samples;

The extraction module 44 is configured to, if the number of positive samples is not less than the preset minimum number of positive samples, and the number of negative samples is not less than the preset minimum number of negative samples, follow the preset total number of positive and negative samples Preset the ratio of positive and negative samples, extract the positive sample image and the negative sample image, and generate a sample training set;

The generating module 45 is configured to train the positive sample image and the negative sample image in the sample training set through the region-based fast convolutional neural network Fast R-CNN model, and generate the weight parameters of the image features of the training image , The image features include shape, color and shadow;

The calculation module 46 is configured to calculate the feature vector value of the image to be tested according to the weight parameter;

The determining module 47 is configured to determine whether the palm image is included in the image to be tested according to the feature vector value.

Further, the marking module 41 includes:

The marking unit 411 is configured to use a solid mask to mark the palm image in the training image;

The obtaining unit 412 is configured to obtain the pixel point coordinates of the marked pixels marked by the solid mask in the training image;

The generating unit 413 is configured to generate a mask matrix according to the pixel coordinates;

The acquiring unit 412 is further configured to perform a dot product operation on the mask matrix and the training image to generate and acquire the palm image.

Further, the extraction module 44 includes:

The saving unit 441 is configured to save the positive sample image in the positive sample library, and save the negative sample image in the negative sample library;

The extraction unit 442 is further configured to extract the positive sample image, the positive sample image, and the negative sample database from the positive sample library and the negative sample library according to preset rules according to the preset total number of positive and negative samples and the ratio of positive and negative samples The negative sample image generates a sample training set.

Further, the generating module 45 includes:

The input unit 451 is used in the Fast In the R-CNN model, input the positive sample image or the negative sample image in the training sample set;

The calculation unit 452 is configured to use a regional candidate network to calculate and generate suggested windows for the number of preset windows of the positive sample image or negative sample image;

The mapping unit 453 is configured to map the suggestion window to the Fast On the feature map convolutional layer of the R-CNN model;

The generating unit 454 is used to combine the feature The map convolution layer and each rectangular ROI of the RoI pooling layer of the Fast R-CNN model perform convolution operations to generate a feature map of a fixed size;

The generating unit 454 is also used to jointly train the classification probability and border regression of the fixed-size feature map to generate the weight parameter of the image feature.

Further, the generating unit 454 is configured to:

Use the detection classification probability Softmax Loss and the detection frame regression Smooth L1 Loss to jointly train the fixed size feature for the classification probability and the frame regression respectively map, generating the weight parameter of the image feature.

Further, the judgment module 47 is configured to:

If the feature vector value is greater than a preset threshold, determining that the image to be tested includes the palm image;

If the feature vector value is not greater than the preset threshold, it is determined that the palm image is not included in the image to be tested.

Further, the method further includes:

The recognition module 48 is configured to recognize palmprint feature information in the palm image if the image to be tested includes the palm image, where the palmprint feature information includes main lines, folds, minutiae points, and triangle points;

The recognition module 48 is also configured to recognize the image to be tested according to the palmprint feature information.

According to an embodiment of the present application, a storage medium is provided, the storage medium stores at least one executable instruction, and the computer executable instruction can execute the palm image detection method in any of the foregoing method embodiments, and the computer readable The storage medium may be non-volatile or volatile.

FIG. 5 shows a schematic structural diagram of a computer device according to an embodiment of the present application, and the specific embodiment of the present application does not limit the specific implementation of the computer device.

As shown in FIG. 5, the computer device may include a processor 502, a communications interface 504, a memory 506, and a communications bus 508.

Wherein: the processor 502, the communication interface 504, and the memory 506 communicate with each other through the communication bus 508.

The communication interface 504 is used to communicate with network elements of other devices, such as clients or other servers.

The processor 502 is configured to execute the program 510, and specifically can execute the relevant steps in the embodiment of the palm image detection method described above.

Specifically, the program 510 may include program code, and the program code includes computer operation instructions.

The processor 502 may be a central processing unit CPU, or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application. The one or more processors included in the computer device may be the same type of processor, such as one or more CPUs; or different types of processors, such as one or more CPUs and one or more ASICs.

The memory 506 is used to store the program 510. The memory 506 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), for example, at least one disk memory.

The program 510 may be specifically used to cause the processor 502 to perform the following operations:

Use a solid mask to mark the palm image in the training image, and obtain the palm image;

Within the image pixel range of the training image, intercept sub-images in the training image according to the image size and pixel position calculated by a random function;

If the sub-image includes the palm image, determine that the sub-image is a positive sample image, and count the number of positive samples of the positive sample image;

If the sub-image does not completely contain the palm image, determining that the sub-image is a negative sample image, and counting the number of negative samples of the negative sample image;

If the number of positive samples is less than the preset minimum number of positive samples, or the number of negative samples is less than the preset minimum number of negative samples, then the sub-images in the training image are captured again;

If the number of positive samples is not less than the preset minimum number of positive samples, and the number of negative samples is not less than the preset minimum number of negative samples, according to the preset total number of positive and negative samples and the preset ratio of positive and negative samples , Extracting the positive sample image and the negative sample image to generate a sample training set;

The positive sample image and the negative sample image in the sample training set are trained by the region-based fast convolutional neural network Fast R-CNN model to generate the weight parameters of the image features of the training image, the image features including Shape, color and shadow;

Calculating the feature vector value of the image to be tested according to the weight parameter;

According to the feature vector value, it is determined whether the image to be tested includes a palm image.

Obviously, those skilled in the art should understand that the above-mentioned modules or steps of this application can be implemented by a general computing device, and they can be concentrated on a single computing device or distributed in a network composed of multiple computing devices. Above, alternatively, they can be implemented with program codes executable by the computing device, so that they can be stored in the storage device for execution by the computing device, and in some cases, can be executed in a different order than here. Perform the steps shown or described, or fabricate them into individual integrated circuit modules, or fabricate multiple modules or steps of them into a single integrated circuit module to achieve. In this way, this application is not limited to any specific hardware and software combination.

The above descriptions are only preferred embodiments of the application, and are not used to limit the application. For those skilled in the art, the application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this application shall be included in the protection scope of this application.

Claims

A method for detecting palm images, including:

Use a solid mask to mark the palm image in the training image, and obtain the palm image;

Within the image pixel range of the training image, intercept sub-images in the training image according to the image size and pixel position calculated by a random function;

If the sub-image includes the palm image, determine that the sub-image is a positive sample image, and count the number of positive samples of the positive sample image;

If the sub-image does not completely contain the palm image, determining that the sub-image is a negative sample image, and counting the number of negative samples of the negative sample image;

If the number of positive samples is less than the preset minimum number of positive samples, or the number of negative samples is less than the preset minimum number of negative samples, then the sub-images in the training image are captured again;

If the number of positive samples is not less than the preset minimum number of positive samples, and the number of negative samples is not less than the preset minimum number of negative samples, according to the preset total number of positive and negative samples and the preset ratio of positive and negative samples , Extracting the positive sample image and the negative sample image to generate a sample training set;

The positive sample image and the negative sample image in the sample training set are trained by the region-based fast convolutional neural network Fast R-CNN model to generate the weight parameters of the image features of the training image, the image features including Shape, color and shadow;

Calculating the feature vector value of the image to be tested according to the weight parameter;

According to the feature vector value, it is determined whether the image to be tested includes the palm image.
The method according to claim 1, wherein said using a solid mask to mark the palm image in the training image and obtaining the palm image comprises:

Using a solid mask to mark the palm image in the training image;

Acquiring the pixel point coordinates of the marked pixels marked by the solid mask in the training image;

Generate a mask matrix according to the pixel coordinates;

Do a dot product operation on the mask matrix and the training image to generate and obtain the palm image.
The method according to claim 1, wherein the extracting the positive sample image and the negative sample image according to the preset total number of positive and negative samples and the preset ratio of positive and negative samples to generate a sample training set comprises:

Save the positive sample image in a positive sample library, and save the negative sample image in a negative sample library;

According to the preset total number of positive and negative samples and the ratio of positive and negative samples, the positive sample image and the negative sample image are respectively extracted from the positive sample library and the negative sample library according to preset rules to generate Sample training set.
The method according to claim 1, wherein the training of the positive sample image and the negative sample image in the sample training set is performed by the region-based fast convolutional neural network Fast R-CNN model to generate the training image The weight parameters of image features include:

In the Fast R-CNN model, input a positive sample image or a negative sample image in the training sample set;

Using the regional candidate network to calculate and generate suggested windows for the number of preset windows for the positive sample image or the negative sample image;

Mapping the suggestion window to the feature map convolutional layer of the Fast R-CNN model;

Performing a convolution operation on the feature map convolution layer and each rectangular ROI of the RoI pooling layer of the Fast R-CNN model to generate a feature map of a fixed size;

Joint training of the classification probability and border regression of the fixed-size feature map to generate the weight parameter of the image feature.
The method according to claim 4, wherein the joint training of the classification probability and border regression of the fixed-size feature map to generate the weight parameter of the image feature comprises:

Use the detection classification probability Softmax Loss and the detection frame regression Smooth L1 Loss to jointly train the fixed size feature for the classification probability and the frame regression respectively map, generating the weight parameter of the image feature.
The method according to claim 1, wherein the judging whether the image to be tested contains the palm image according to the feature vector value comprises:

If the feature vector value is greater than a preset threshold, determining that the image to be tested includes the palm image;

If the feature vector value is not greater than the preset threshold, it is determined that the palm image is not included in the image to be tested.
7. The method according to claim 6, wherein after said determining whether the image to be tested contains the palm image according to the feature vector value, the method further comprises:

If the image to be tested includes the palm image, identifying palmprint feature information in the palm image, where the palmprint feature information includes main lines, folds, minutiae points, and triangle points;

Identify the image to be tested according to the palmprint feature information.
A palm image detection device, which includes:

The marking module is used to mark the palm image in the training image with a solid mask and obtain the palm image;

The interception module is used to intercept the sub-images in the training image according to the image size and pixel position calculated by the random function within the image pixel range of the training image;

A determining module, configured to determine that the sample image is a positive sample image if the sub-image includes the palm image, and count the number of positive samples of the positive sample image;

The determining module is further configured to determine that the sample image is a negative sample image if the sub-image does not completely include the palm image, and count the number of negative samples of the negative sample image;

The intercepting module is configured to intercept the sub-image in the training image again if the number of positive samples is less than the preset minimum number of positive samples, or the number of negative samples is less than the preset minimum number of negative samples;

The extraction module is configured to, if the number of positive samples is not less than the preset minimum number of positive samples, and the number of negative samples is not less than the preset minimum number of negative samples, follow the preset total number of positive and negative samples and the preset Set the ratio of positive and negative samples, extract the positive sample image and the negative sample image, and generate a sample training set;

Generation module, used to pass region-based fast convolutional neural network Fast The R-CNN model trains the positive sample image and the negative sample image in the sample training set, and generates weight parameters of image features of the training image, the image features including shape, color, and shadow;

The calculation module is used to calculate the feature vector value of the image to be tested according to the weight parameter;

The judging module is used to judge whether the image to be tested contains a palm image according to the feature vector value.
A storage medium storing at least one executable instruction, and the executable instruction causes a processor to perform the operations of the following steps:

Use a solid mask to mark the palm image in the training image, and obtain the palm image;

Within the image pixel range of the training image, intercept sub-images in the training image according to the image size and pixel position calculated by a random function;

If the sub-image includes the palm image, determine that the sub-image is a positive sample image, and count the number of positive samples of the positive sample image;

If the sub-image does not completely contain the palm image, determining that the sub-image is a negative sample image, and counting the number of negative samples of the negative sample image;

If the number of positive samples is less than the preset minimum number of positive samples, or the number of negative samples is less than the preset minimum number of negative samples, then the sub-images in the training image are captured again;

If the number of positive samples is not less than the preset minimum number of positive samples, and the number of negative samples is not less than the preset minimum number of negative samples, according to the preset total number of positive and negative samples and the preset ratio of positive and negative samples , Extracting the positive sample image and the negative sample image to generate a sample training set;

Through the region-based fast convolutional neural network Fast The R-CNN model trains the positive sample image and the negative sample image in the sample training set, and generates weight parameters of image features of the training image, the image features including shape, color, and shadow;

Calculating the feature vector value of the image to be tested according to the weight parameter;

According to the feature vector value, it is determined whether the image to be tested includes the palm image.
9. The storage medium of claim 9, wherein the step of using a solid mask to mark the palm image in the training image and obtaining the palm image comprises:

Using a solid mask to mark the palm image in the training image;

Acquiring the pixel point coordinates of the marked pixels marked by the solid mask in the training image;

Generate a mask matrix according to the pixel coordinates;

Do a dot product operation on the mask matrix and the training image to generate and obtain the palm image.
9. The storage medium of claim 9, wherein the extracting the positive sample image and the negative sample image according to the preset total number of positive and negative samples and the preset ratio of positive and negative samples to generate a sample training set comprises:

Save the positive sample image in a positive sample library, and save the negative sample image in a negative sample library;

According to the preset total number of positive and negative samples and the ratio of positive and negative samples, the positive sample image and the negative sample image are respectively extracted from the positive sample library and the negative sample library according to preset rules to generate Sample training set.
The storage medium according to claim 9, wherein the training of the positive sample image and the negative sample image in the sample training set is performed by the region-based fast convolutional neural network Fast R-CNN model to generate the training The weight parameters of the image features of the image include:

In the Fast R-CNN model, input a positive sample image or a negative sample image in the training sample set;

Using the regional candidate network to calculate and generate suggested windows for the number of preset windows for the positive sample image or the negative sample image;

Mapping the suggestion window to the feature map convolutional layer of the Fast R-CNN model;

Performing a convolution operation on the feature map convolution layer and each rectangular ROI of the RoI pooling layer of the Fast R-CNN model to generate a feature map of a fixed size;

Joint training of the classification probability and border regression of the fixed-size feature map to generate the weight parameter of the image feature.
9. The storage medium according to claim 9, wherein the determining whether the image to be tested contains the palm image according to the feature vector value comprises:

If the feature vector value is greater than a preset threshold, determining that the image to be tested includes the palm image;

If the feature vector value is not greater than the preset threshold, it is determined that the palm image is not included in the image to be tested.
15. The storage medium according to claim 13, wherein, after said determining whether the image to be tested includes the palm image according to the feature vector value, the method further comprises:

If the image to be tested includes the palm image, identifying palmprint feature information in the palm image, where the palmprint feature information includes main lines, folds, minutiae points, and triangle points;

Identify the image to be tested according to the palmprint feature information.
A computer device includes: a processor, a memory, a communication interface, and a communication bus. The processor, the memory, and the communication interface communicate with each other through the communication bus;

The memory is used to store at least one executable instruction, and the executable instruction causes the processor to perform the operations of the following steps:

Use a solid mask to mark the palm image in the training image, and obtain the palm image;

Within the image pixel range of the training image, intercept sub-images in the training image according to the image size and pixel position calculated by a random function;

If the sub-image includes the palm image, determine that the sub-image is a positive sample image, and count the number of positive samples of the positive sample image;

If the sub-image does not completely contain the palm image, determining that the sub-image is a negative sample image, and counting the number of negative samples of the negative sample image;

If the number of positive samples is less than the preset minimum number of positive samples, or the number of negative samples is less than the preset minimum number of negative samples, then the sub-images in the training image are captured again;

If the number of positive samples is not less than the preset minimum number of positive samples, and the number of negative samples is not less than the preset minimum number of negative samples, according to the preset total number of positive and negative samples and the preset ratio of positive and negative samples , Extracting the positive sample image and the negative sample image to generate a sample training set;

Through the region-based fast convolutional neural network Fast The R-CNN model trains the positive sample image and the negative sample image in the sample training set, and generates weight parameters of image features of the training image, the image features including shape, color, and shadow;

Calculating the feature vector value of the image to be tested according to the weight parameter;

According to the feature vector value, it is determined whether the image to be tested includes the palm image.
15. The computer device according to claim 15, wherein said using a solid mask to mark a palm image in a training image and obtaining said palm image comprises:

Using a solid mask to mark the palm image in the training image;

Acquiring the pixel point coordinates of the marked pixels marked by the solid mask in the training image;

Generate a mask matrix according to the pixel coordinates;

Do a dot product operation on the mask matrix and the training image to generate and obtain the palm image.
15. The computer device according to claim 15, wherein said extracting said positive sample image and said negative sample image according to the preset total number of positive and negative samples and the preset ratio of positive and negative samples to generate a sample training set comprises:

Save the positive sample image in a positive sample library, and save the negative sample image in a negative sample library;

According to the preset total number of positive and negative samples and the ratio of positive and negative samples, the positive sample image and the negative sample image are respectively extracted from the positive sample library and the negative sample library according to preset rules to generate Sample training set.
The computer device according to claim 15, wherein the positive sample image and the negative sample image in the sample training set are trained by the region-based fast convolutional neural network Fast R-CNN model to generate the training image The weight parameters of image features include:

In the Fast R-CNN model, input a positive sample image or a negative sample image in the training sample set;

Using the regional candidate network to calculate and generate suggested windows for the number of preset windows for the positive sample image or the negative sample image;

Mapping the suggestion window to the feature map convolutional layer of the Fast R-CNN model;

Performing a convolution operation on the feature map convolution layer and each rectangular ROI of the RoI pooling layer of the Fast R-CNN model to generate a feature map of a fixed size;

Joint training of the classification probability and border regression of the fixed-size feature map to generate the weight parameter of the image feature.
The computer device according to claim 15, wherein the determining whether the image to be tested contains the palm image according to the feature vector value comprises:

If the feature vector value is greater than a preset threshold, determining that the image to be tested includes the palm image;

If the feature vector value is not greater than the preset threshold, it is determined that the palm image is not included in the image to be tested.
The computer device according to claim 19, wherein after said determining whether the image to be tested contains the palm image according to the feature vector value, the method further comprises:

If the image to be tested includes the palm image, identifying palmprint feature information in the palm image, where the palmprint feature information includes main lines, folds, minutiae points, and triangle points;

Identify the image to be tested according to the palmprint feature information.