Biological identity recognition method and system and computer equipment

Abstract

The embodiment of the invention relates to the technical field of biological identity recognition, in particular to a biological identity recognition method and system and computer equipment. The method mainly comprises the steps of performing foreground extraction on an obtained initial image of a to-be-recognized organism to obtain a foreground image corresponding to a target area in the initial image, performing feature enhancement processing on the foreground image to obtain a feature-enhanced target foreground image, performing similarity comparison on the target foreground image and a plurality of sample images, and obtaining a feature-enhanced target foreground image; and determining target identity information of the to-be-recognized organism corresponding to the target foreground image. By adopting the method, the accuracy of biological identity recognition can be improved.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of biological identity recognition, and in particular to a biological identity recognition method, system and computer device. BACKGROUND

[0002] With the development of image recognition technology, accurate recognition of biological identity in images has become a core technical support for scenarios such as rare species protection, wild animal population dynamic monitoring, and endangered biological individual tracking. Typical applications include snow leopard individual identification in nature reserves, cetacean growth state tracking, and wild bird population migration monitoring.

[0003] Current mainstream biological identity recognition technologies mainly include two categories: one is automatic recognition based on pre-trained artificial intelligence models, and the other is manual comparison and judgment relying on expert experience. However, both have obvious limitations: at the model training level, high-precision recognition requires massive sample data, but image data of rare and endangered species is generally scarce, making it difficult to meet training needs, which limits the accuracy of model recognition; at the manual comparison level, limited by labor costs and subjective judgment efficiency, the overall comparison efficiency is low, making it difficult to adapt to timely and efficient comparison needs. SUMMARY

[0004] Therefore, it is necessary to provide a biological identity recognition method, system and computer device that improves recognition accuracy and efficiency to solve at least one of the above technical problems.

[0005] In a first aspect, embodiments of the present disclosure provide a biological identity recognition method, which can include the following steps: performing foreground extraction on an obtained initial image of a to-be-identified biological object to obtain a foreground image corresponding to a target region in the initial image, performing feature enhancement processing on the foreground image to obtain a target foreground image after feature enhancement, performing similarity comparison between the target foreground image and a plurality of sample images, and determining target identity information of the to-be-identified biological object corresponding to the target foreground image.

[0006] The target region includes a target object on the surface of the to-be-identified biological object, and the target object includes at least one of a spot, a texture, and a part with biological identity characteristics. The sample image is labeled with identity information of a corresponding biological object.

[0007] In some embodiments, performing foreground extraction on the obtained initial image of the to-be-identified biological object to obtain a foreground image corresponding to a target region in the initial image can include the following steps: performing instance segmentation on the initial image using a pre-set deep learning network to obtain preliminary foreground information, and performing visual processing on the preliminary foreground information to obtain the foreground image.

[0008] In some embodiments, determining the target identity information of the to-be-identified biological object corresponding to the target foreground image based on the similarity comparison between the target foreground image and the plurality of sample images can include the following steps: calculating similarity scores of the target foreground image and the plurality of sample images according to a preset similarity evaluation index; and determining a target sample image from the plurality of sample images according to the similarity scores, and taking identity information of the target sample image as the target identity information of the target foreground image.

[0009] In some embodiments, the similarity evaluation index includes a structural similarity index, a two-dimensional correlation coefficient, a mean square error value, and a feature point matching distance, and the target sample image and the target foreground image are images of the same biological object at different times.

[0010] In some embodiments, determining the target sample image from the plurality of sample images according to the similarity scores can include the following steps: determining a plurality of candidate sample images from the plurality of sample images according to the similarity scores; sending the plurality of candidate sample images and the target foreground image to a user terminal; and determining the target sample image according to a feedback confirmation result of the user terminal.

[0011] In some embodiments, calculating the similarity scores of the target foreground image and the plurality of sample images according to the preset similarity evaluation index can include the following steps: extracting image features of the target region to obtain a target feature set; extracting image features of the plurality of sample images to obtain a plurality of sample feature sets respectively corresponding to the plurality of sample images; performing feature quantization processing on the target feature set and the plurality of sample feature sets to obtain a target feature vector and a plurality of sample feature vectors; and calculating the target feature vector and the plurality of sample feature vectors according to the preset similarity index to obtain a plurality of similarity scores respectively corresponding to the plurality of sample images.

[0012] In some embodiments, the image features include geometric properties of the target object in the target region.

[0013] In some embodiments, the biological identity recognition method can further include the following steps: obtaining a group image to be identified; performing foreground extraction on the group image to obtain a plurality of group foreground images respectively corresponding to a plurality of to-be-identified biological objects; adding a group label to the plurality of group foreground images; and when any group foreground image is a special group foreground image without a target object, identity information of the to-be-identified biological object corresponding to the special group foreground image includes the group label.

[0014] In some embodiments, the group image includes a plurality of to-be-identified biological objects, and the group label is used to represent a group relationship between the plurality of to-be-identified biological objects.

[0015] In a second aspect, embodiments of the present disclosure provide a biological identity recognition system, which can include a processor, a user terminal, and a sample database.

[0016] The processor is configured to acquire an initial image of the to-be-identified living body from the user terminal, and perform the steps of the living body identity recognition method provided in any of the embodiments of the first aspect.

[0017] The sample database is configured to store a plurality of sample images, and the sample images are marked with identity information of corresponding living bodies.

[0018] In some embodiments, the processor is further configured to perform similarity comparison between the target foreground image and the plurality of sample images, calculate similarity scores of the target foreground image and the plurality of sample images, and determine a plurality of candidate sample images from the plurality of sample images according to the similarity scores, and send the plurality of candidate sample images and the target foreground image to the user terminal.

[0019] The user terminal is configured to visually present the plurality of candidate sample images and the target foreground image, and feed back a confirmation result to the processor.

[0020] The processor is further configured to determine a target sample image according to the confirmation result, and take identity information of the target sample image as target identity information of the target foreground image.

[0021] In a third aspect, an embodiment of the present disclosure provides a computer device, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the living body identity recognition method provided in any of the embodiments of the first aspect when executing the computer program.

[0022] In a fourth aspect, an embodiment of the present disclosure provides a computer-readable storage medium, which stores a computer program, and the computer program is executable on a processor to implement the steps of the living body identity recognition method provided in any of the embodiments of the first aspect.

[0023] The aforementioned biometric identification method, system, computer equipment, and storage medium extract foreground features from the initial image of the organism to be identified, enhance the features in the foreground image to highlight the detailed information of the corresponding features of the target object, and obtain a target foreground image. The target foreground image is then compared with sample images to determine the target identity information of the organism corresponding to the target foreground image. This method has the advantages of high accuracy and high efficiency. Furthermore, the identification of the organism can be based on at least one of the following: spots, textures, or specific body parts, increasing the flexibility of the feature requirements for the initial image of the organism to be identified. Compared to manual comparison and judgment methods that rely entirely on expert experience, this identification method achieves automated identification through a standardized algorithm process, thus improving identification efficiency. Compared to model-based identification methods, it does not require massive amounts of sample data to complete the identification, reducing dependence on the scale of sample data. At the same time, this method adopts a "white-box" identification approach, which has strong interpretability and traceability of the processing process. Key parameters can be adjusted in reverse based on actual identification results to optimize performance, and the model does not need to be retrained when new organisms are identified. Attached Figure Description

[0024] Figure 1 This is a diagram illustrating the application environment of biometric identification methods in some embodiments; Figure 2 This is a flowchart illustrating the biometric identification method in some embodiments; Figure 3 This is a flowchart illustrating the foreground extraction step in some embodiments; Figure 4 This is a flowchart illustrating the steps involved in determining the target identity information in some embodiments; Figure 5 This is a flowchart illustrating the target sample image determination step in some embodiments; Figure 6 This is a flowchart illustrating the steps involved in calculating similarity scores in some embodiments; Figure 7 This is a flowchart illustrating the group image recognition steps in some embodiments; Figure 8 Here are some structural block diagrams of biometric identification systems in various embodiments; Figure 9 This is a diagram showing the internal structure of a computer device in some embodiments. Detailed Implementation

[0025] To make the technical solutions and advantages of this disclosure clearer, the embodiments and related technical content of this disclosure will be further described in detail below with reference to the accompanying drawings and text description. It should be understood that the embodiments described below are only used to explain the technical solutions of the embodiments of this disclosure and are not intended to limit more possible implementations of this disclosure.

[0026] It should be noted that relational terms such as "first" and "second" appearing in this document are used only to distinguish things, states, or actions, and do not necessarily indicate or imply relative importance or order. The terms "including," "comprising," or any other variations thereof are used to indicate non-exclusive inclusion, and the included objects may not be limited to those listed in this document. The terms "multiple" or other variations are used to indicate that the number of objects is two or more.

[0027] In a first aspect, embodiments of this disclosure provide a biometric identification method. This method can be applied to, for example... Figure 1 In the application environment shown, server 101 can communicate with database 102 and terminal 103 via a network. Server 101 can be implemented as a standalone server or a server cluster consisting of multiple servers. Database 102 can be a local database or a cloud database, used to store sample images of multiple organisms. Terminal 103 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices.

[0028] The processor in server 101 can be implemented using at least one of the following hardware forms: programmable logic array (PLA), field-programmable gate array (FPGA), digital signal processor (DSP), application-specific integrated circuit (ASIC), general-purpose processor, or other programmable logic device.

[0029] Applying biometric identification methods Figure 1 Taking server 101 as an example, in some embodiments, such as Figure 2 As shown, the biometric identification method includes steps S201 to S203 that can be executed by server 101. Each step is described in detail below.

[0030] Step S201: Extract the foreground from the initial image of the organism to be identified to obtain the foreground image corresponding to the target region in the initial image.

[0031] The initial image can be an image including a full view or a partial view of the biological individual to be identified, can be a photo or a video frame of the biological individual to be identified. The target region includes a target object on the body surface of the biological individual to be identified, that is, the target region is a region corresponding to the target object on the body surface of the biological individual to be identified. The target object includes at least one of a spot, a texture, and a part with biological identity characteristics. For example, when the biological individual to be identified is a cetacean (e.g., a whale, a dolphin, etc.), the target object can be at least one of a spot, a texture (e.g., a wound, a scar, a notch), and a dorsal fin or a tail fin on the body surface of the cetacean. For another example, when the biological individual to be identified is a feline (e.g., a tiger, a leopard, a lion, etc.), the target object can be at least one of a spot (a color block of body hair), a texture (e.g., a pattern), or a face (e.g., an eye, a nose, a forehead) on the body surface of the feline.

[0032] In some embodiments, as shown in FIG. 3, step S201 can include step S301 and step S302. Figure 3

[0033] Step S301: performing instance segmentation on the initial image by using a preset deep learning network to obtain preliminary foreground information. Step S302: performing visual processing on the preliminary foreground information to obtain a foreground image.

[0034] The preset deep learning network can be UNet (U-Shaped Convolutional Neural Network), Mask R-CNN (Mask Region-based Convolutional Neural Network), SOLO (Segmenting Objects by Locations), or MVANet (Multi-View Analysis Network), IndexNet Matting (Index Convolutional Neural Network for Image Matting), etc., which can be used to perform foreground extraction.

[0035] The preset deep learning network can be a pre-trained foreground extraction model. By inputting the initial image into the foreground extraction model, preliminary foreground information after instance segmentation can be obtained.

[0036] ​Specifically, the content of the preliminary foreground information output by different types of deep learning networks can be different. When the deep learning network is UNet or MVANet, the preliminary foreground information obtained is a mask, and each pixel value represents the probability that the position belongs to the foreground (1) or the background (0), or a binary mask is directly output. When the deep learning network is MVANet or Mask R-CNN, the preliminary foreground information obtained can include a target bounding box, a class label, and a mask corresponding to an instance.

[0037] The preliminary foreground information can be visualized by performing mask processing and mask superimposition to obtain a foreground image. Specifically, the probability mask output by the deep learning network is binarized, for example, by setting a threshold value of 0.5, and a value greater than the threshold value is foreground and otherwise is background. If it is instance segmentation, the mask of a specific class can be selected according to requirements. The mask is multiplied with the original image at the pixel level to filter out background pixels for mask superimposition. After completing the mask superimposition processing, a post-processing step can also be included: converting the color space of the foreground image from RGB (a color standard) to Lab (a color model), segmenting the a channel, and adjusting the size and grid of the foreground image.

[0038] Specifically, the post-processing step can include color space conversion, edge smoothing processing, size adjustment, and / or format conversion. For example, to optimize the mask, the initial image or the preliminary foreground information output by the neural network model (such as the initial mask) can be converted from RGB to Lab format, the a channel is segmented, and morphological processing such as opening operation or closing operation is performed; combined with equal scale size adjustment (such as equal scale enlargement to occupy the full frame), the final foreground image after post-processing is obtained. The output format of the foreground image can be PNG (Portable Network Graphics, a bitmap format using lossless compression algorithm) or JPG (Joint Photographic Experts Group, a compression standard for continuous-tone still images) format.

[0039] In some specific examples, step S201 can further include extracting according to a preset foreground extraction parameter to obtain the foreground image. Specifically, the preset foreground extraction parameter can include a grayscale threshold or a color threshold, and threshold segmentation is performed according to the grayscale threshold and the color threshold to obtain the foreground image.

[0040] In some specific examples, step S201 can further include extracting the image edge of the initial image according to a preset edge detection algorithm to obtain image edge information, obtaining a foreground contour according to the image edge information, and obtaining the foreground image according to the foreground contour.

[0041] It is easy to understand that for the foreground extraction manner of step S201, other manners can also be adopted, such as region growing-based foreground extraction or background modeling-based foreground extraction manner, etc.

[0042] In some specific examples, the biological identity recognition method further includes the following steps: acquiring an initial image of the biological to be recognized. Specifically, the initial image of the biological to be recognized corresponding to the identity recognition instruction can be acquired in response to the identity recognition instruction initiated by the terminal 103. Wherein, the terminal 103 initiates the identity recognition instruction at the same time, and the initial image of the biological to be recognized is uploaded together.

[0043] Step S202: performing feature enhancement processing on the foreground image to obtain a target foreground image after feature enhancement.

[0044] In some specific examples, the feature enhancement processing on the foreground image can adopt one or more of the following manners: wavelet transform processing, automatic adjustment of histogram, and adaptive threshold adjustment, etc.

[0045] Taking the wavelet transform processing as an example, the wavelet transform processing on the foreground image can include the following steps: image preprocessing, multi-scale wavelet decomposition, high-frequency coefficient enhancement processing, and wavelet inverse transform reconstruction, etc. Wherein, the image preprocessing step can be normalization processing on the foreground image, mapping the pixel value to a specific interval, avoiding the influence of the range difference of the pixel value on the transformation effect, and if the foreground image is color, the preprocessing can be converting the color image to a grayscale image. The foreground image after preprocessing is decomposed and high-frequency coefficient enhanced according to the preset wavelet basis function and decomposition level, to obtain high-frequency coefficients and low-frequency components, and the high-frequency coefficients and low-frequency components are inverse transformed to obtain the target foreground image after feature enhancement.

[0046] Through the feature enhancement processing on the foreground image, the features of the target object in the image can be highlighted and enhanced.

[0047] Step S203: comparing the target foreground image with multiple sample images to determine the target identity information of the biological to be recognized corresponding to the target foreground image.

[0048] Wherein, the sample images are marked with the identity information of the corresponding biological. The sample images can be historical images with biological identity information stored in the database 102. The sample images of the same biological can include multiple historical images of different angles and different contents. In some specific examples, the sample images can be biological original images or video frames taken in a historical period, and can also be sample foreground images obtained by performing foreground extraction and feature enhancement processing on the biological original images or video frames.

[0049] Specifically, the target foreground image and the plurality of sample images are compared in similarity, the target foreground image and the plurality of sample images are compared one by one to obtain a similarity score of each sample image and the target foreground image, the target sample image of the same biological object as the target foreground image is determined according to the similarity score, and the identity information of the target sample image is taken as the target identity information of the target foreground image.

[0050] The similarity comparison can be performed by a structural similarity comparison method, a feature point matching method, a pixel direct comparison method or other similarity comparison methods.

[0051] In some specific examples, when the sample image is a biological original image or a video frame, the similarity comparison between the target foreground image and the plurality of sample images can include foreground extraction and wavelet transform processing of the sample image to obtain a sample foreground image, similarity comparison between the target foreground image and the plurality of sample foreground images to determine the target identity information of the corresponding biological object to be identified.

[0052] Taking a cetacean as an example, in the related art, the biological identity recognition method needs to rely on a specific anatomical region corresponding to the cetacean, for example, the individual identity recognition of a whale can only be based on the shape of the tail fin, and the dorsal fin shape of a white dolphin. Such recognition method needs to be based on clear specific local images of the biological object to be identified. However, the image acquisition of the cetacean is often affected by the sea surface light, spray, shooting angle and other factors, and it is difficult to accurately collect clear images of the specific anatomical region, resulting in the problems of high recognition difficulty and low recognition accuracy. At the same time, most of the recognition methods in the related art use specific key point detection or manual annotation of specific region extraction, which is difficult to adapt to the changes of biological individual posture, occlusion and local invisibility in natural environment, and has the problem of poor flexibility of recognition method.

[0053] In the above biological identity recognition method, the initial image of the to-be-recognized biological is subjected to foreground extraction, and the features in the foreground image are enhanced, highlighting the detailed information of the features corresponding to the target object, obtaining a target foreground image, and then comparing the target foreground image with the sample image to determine the target identity information of the to-be-recognized biological corresponding to the target foreground image, which has the advantages of high recognition accuracy and high recognition efficiency. The recognition of the identity of the to-be-recognized biological can be based on at least one of the spots, textures or specific parts on the body surface, and can recognize biological images containing the whole body, as well as biological images including only local regions of the individual, and can adapt to various biological images collected under conditions of individual posture changes, occlusions and local invisibility. The flexibility of the feature requirements of the initial image of the to-be-recognized biological is improved. Compared with the manual comparison and judgment method which completely relies on expert experience, the recognition method realizes automatic recognition through a standardized algorithm process, which can improve the recognition efficiency. Compared with the model recognition method, the recognition can be completed without relying on a large amount of sample data, reducing the dependence on the size of sample data. At the same time, the method adopts a "white box" type of recognition, has strong interpretability, the processing process is traceable, and the key parameters can be adjusted reversely according to the actual recognition effect to optimize the performance. When a new biological individual is identified, the model does not need to be retrained.

[0054] In some embodiments, as shown in FIG. 4, step S203 can include steps S401-S403. Figure 4

[0055] Step S401: According to a preset similarity evaluation index, the similarity scores of the target foreground image and the plurality of sample images are calculated.

[0056] The similarity evaluation index includes a structural similarity index, a two-dimensional correlation coefficient, a mean square error value, and a feature point matching distance.

[0057] Specifically, the similarity scores of the target foreground image and the plurality of sample images can be calculated based on the respective similarity evaluation indexes. That is, for each sample image, there are a plurality of similarity scores corresponding to the respective similarity evaluation indexes.

[0058] Step S402: According to the similarity scores, the target sample image is determined from the plurality of sample images.

[0059] The target sample image and the target foreground image are images of the same biological at different times.

[0060] In some specific embodiments, as shown in FIG. 5, step S402 can include steps S501-S503. Figure 5

[0061] ​​Step S501: determining a plurality of candidate sample images from the plurality of sample images according to the similarity scores.

[0062] In some specific examples, the candidate sample image with the highest similarity score can be selected from the plurality of sample images corresponding to each similarity evaluation index, and then the plurality of samples are obtained. Specifically, the similarity scores of different sample images calculated based on each similarity evaluation index are shown in Table 1.

[0063] Table 1: Similarity score table ; That is, the sample image with the highest score can be selected from to . The sample image with the highest score can be selected from to . The sample image with the highest score can be selected from to . The sample image with the highest score can be selected from to . The sample images corresponding to , , and are respectively taken as candidate sample images. In some cases, the candidate sample images obtained according to different similarity evaluation indexes can be the same sample image.

[0064] In yet some specific examples, the comprehensive similarity scores can also be obtained by performing weighted calculation according to the plurality of similarity scores of each sample image corresponding to a plurality of similarity evaluation indexes, and the sample images in the front of the ranking are taken as candidate sample images according to the comprehensive similarity scores of the sample images. For example, the top 3 sample images in the ranking can be taken as candidate sample images, and the number of selected candidate sample images can be set by a person skilled in the art according to actual needs.

[0065] In yet some examples, step S501 can further include: determining a plurality of candidate sample images from the plurality of sample images according to the similarity scores and a preset sample threshold.

[0066] The similarity score of the candidate sample image is greater than the preset sample threshold. The preset sample threshold is used to represent the similarity degree. When the similarity score of any sample image and the target foreground image is greater than the preset sample threshold, it indicates that the similarity degree of the sample image and the target foreground image is high.

[0067] Step S502: sending the plurality of candidate sample images and the target foreground image to the user end.

[0068] Specifically, the plurality of candidate sample images and the target foreground image are sent to a user terminal (e.g., terminal 103). The terminal 103 is configured to present the candidate sample images one by one in pairs with the target foreground image in a preset visualization manner. The preset visualization manner can include, for example, superimposed comparison, feature point connection comparison, flashing switching comparison, local highlighting comparison, color mapping, and the like.

[0069] The superimposed comparison is to superimpose the candidate sample images and the contour lines of the to-be-recognized biometric feature points in the target foreground image, so that the staff can clearly see the feature shape contour differences of the comparison group from the display interface of the terminal 103.

[0070] The feature point connection comparison is to connect the feature points in the candidate sample images one by one, and also connect the feature points in the target foreground image one by one, so that the staff can judge the shape of the feature point connection.

[0071] The flashing switching comparison is to quickly and alternately display the comparison group images on the display interface of the terminal 103, so that the staff can capture the differences between the two images.

[0072] The local highlighting comparison is to generate a difference image by calculating the pixel difference between the two images, and then highlight the inconsistent areas in the difference image using a heat map, so as to assist the staff in focusing on the local difference features. The local highlighting comparison can also be implemented by generating a sliding block or a mask on the display interface of the terminal 103, and the staff can control the movement of the blocking component based on an input device to highlight the local area for identification. In some specific examples, the blocking component can be an irregular mask that extends from the edge of the foreground region in the target foreground image to the background region, and covers the background region. The coverage range of the irregular mask can be dynamically adjusted according to the input instructions (e.g., mouse dragging or keyboard operation) of the staff.

[0073] The color mapping is to use different color channels (e.g., red / blue) for the comparison group images, and then reflect the corresponding point displacement by the color deviation after superimposition.

[0074] In some specific examples, the preset visualization manner can also include a deformation heat map manner and a side-by-side comparison manner. The deformation heat map manner is to highlight the stretched or compressed areas after affine transformation by the color gradient of the heat map. The side-by-side comparison manner is to display the comparison group images side by side.

[0075] The preset visual effects can be combined in one or more manners in actual application. By presenting the target foreground image and the candidate sample image in a visual manner, the identification efficiency of the staff can be improved.

[0076] Step S503: determining the target sample image according to the confirmation result fed back by the user end.

[0077] The confirmation result fed back by the user end can be a confirmation result automatically generated by the terminal 103 according to the point selection operation of the staff on the display interface of the terminal 103, wherein the confirmation result includes the identity information of the target sample image.

[0078] By adding the artificial confirmation link, the accuracy of the identification result can be improved.

[0079] Step S403: taking the identity information of the target sample image as the target identity information of the target foreground image.

[0080] The target sample image and the target foreground image are different images of the same biological object, and the identity recognition result of the biological object corresponding to the target foreground image is the identity information of the target sample image.

[0081] In some embodiments, as Figure 6 Step S401 can include steps S601 to S604.

[0082] Step S601: extracting image features of the target region to obtain a target feature set.

[0083] Step S602: extracting image features of multiple sample images respectively to obtain multiple sample feature sets respectively corresponding to the sample images.

[0084] Step S603: performing feature quantization processing on the target feature set and the multiple sample feature sets to obtain a target feature vector and multiple sample feature vectors.

[0085] Step S604: calculating the target feature vector and the multiple sample feature vectors according to a preset similarity index to obtain multiple similarity scores respectively corresponding to the sample images.

[0086] The image features include geometric properties of the target object in the target region. The geometric properties include boundary shape features of the target object, such as shape, perimeter, area, and edge smoothness of spots / textures / positions.

[0087] When the target objects in the target region further include spots and / or textures, the image features can further include a number of the target objects, such as a number of the spots and a number of the textures; a distribution pattern between the target objects, such as a relative position (distance, angle) / arrangement rule between the spots and the spots, a relative position / arrangement rule between the spots and the textures, and a relative position / arrangement rule between the spots / textures and the parts; connectivity between the target objects, such as a specific shape after connectivity between adjacent target objects; a combined shape between the target objects, such as an outline shape formed by a plurality of target objects; a positional relationship, such as a topological structure of the target objects and a distribution characteristic of the centers of gravity; and a texture / brightness feature, such as a gray or color gradient in the spots, an edge contrast of the spots, and the like.

[0088] In some specific examples, the image features of the target region can be geometric attribute features or other image features of all the target objects in the target region, to obtain a target feature set.

[0089] The image features of the plurality of sample images can be image features of the target regions in the sample images, which can be the same type of features as the image features of the target regions extracted from the target foreground image, such as geometric attribute features. The image features of the plurality of sample images are extracted to obtain a plurality of sample feature sets corresponding to the plurality of sample images, respectively.

[0090] The target feature set and the plurality of sample feature sets are subjected to feature quantization processing, respectively, to obtain a target feature vector and a plurality of sample feature vectors. The target feature vector corresponds to the target feature set, and the plurality of sample feature vectors correspond to the plurality of sample feature sets, respectively.

[0091] The target feature vector and the plurality of sample feature vectors are combined into a plurality of control groups, and the number of the control groups is consistent with the number of the sample feature vectors. Similarity scores of the control groups are calculated according to a preset similarity index, to obtain a plurality of similarity scores corresponding to the plurality of sample images.

[0092] In some embodiments, the step S401 of calculating the similarity scores can further include: matching the plurality of sample images with the target foreground image into a plurality of control groups, respectively; constructing a point-to-point distance matrix for feature points of the target objects in each control group to obtain a plurality of distance matrices; and calculating similarity scores corresponding to the plurality of distance matrices, respectively.

[0093] Specifically, the similarity scores corresponding to the plurality of distance matrices can be calculated by an aggregation method, such as an average nearest neighbor distance, a Hausdorff distance, or a bidirectional matching rate.

[0094] It is easy to understand that when the sample image is a biological original image or a video frame taken in a historical period, a foreground extraction operation needs to be performed on the sample image before the similarity comparison. The foreground extraction method for the sample image is consistent with the foreground extraction method for the initial image of the to-be-identified biological.

[0095] In some embodiments, the biological identity recognition method further includes: after the identity information of the to-be-identified biological corresponding to the initial image is determined, storing the initial image or the foreground image corresponding to the initial image as a new sample image in the database 102.

[0096] In some embodiments, the biological identity recognition method can further include the following steps: performing foreground extraction on the newly acquired initial image of the to-be-identified biological to obtain a new foreground image; According to the preset similarity evaluation index, the similarity scores of the new initial image and the plurality of sample images are calculated to obtain a plurality of similarity scores corresponding to the plurality of sample images. When each similarity score is less than the preset sample threshold, the new initial image or the new foreground image is labeled with new identity information, and the new initial image or the new foreground image is stored in the database 102.

[0097] The new identity information has a unique identity identification function, and can distinguish the identity of the biological corresponding to the new initial image or the new foreground image from the identities of other biological corresponding to the sample images in the database.

[0098] Each similarity score being less than the preset sample threshold indicates that the biological corresponding to the new initial image is not recorded in the database 102, and is a new biological individual image.

[0099] In some embodiments, as shown in FIG. 7, Figure 7 The biological identity recognition method can further include steps S701 to S704.

[0100] Step S701: acquiring a group image to be identified.

[0101] Step S702: performing foreground extraction on the group image to obtain a plurality of group foreground images corresponding to a plurality of to-be-identified biologies, respectively.

[0102] Step S703: adding a group label to the plurality of group foreground images.

[0103] Step S704: when any group foreground image is a special group foreground image without a target object, the identity information of the to-be-identified biological corresponding to the special group foreground image includes the group label.

[0104] The group image includes a plurality of to-be-identified organisms, and the group mark is used to represent the group relationship between the plurality of to-be-identified organisms. The group mark can be in the form of digital coding or letter coding.

[0105] In some scenarios, the image to be identified includes a plurality of to-be-identified organisms, that is, the organisms in the image appear in the form of a group. For example, young dolphins usually follow adult dolphins (for example, dolphin female parents) to appear, and there is a possibility of being photographed in the same image. Alternatively, a plurality of dolphins with a blood relationship appear at the same time and are photographed in the same image. Such an image with a plurality of to-be-identified organisms is a group image.

[0106] Specifically, the biological identification method further includes obtaining the group image to be identified, performing foreground extraction on the group image to obtain a plurality of group foreground images corresponding to the plurality of to-be-identified organisms respectively, each group foreground image corresponding to a to-be-identified organism, and adding a group mark to the plurality of group foreground images. When any group foreground image is a special group foreground image without a target object, the identity information of the to-be-identified organism corresponding to the special group foreground image includes the group mark. That is, the image region of the plurality of to-be-identified organisms included in the group image does not have the feature information corresponding to the target object. For example, the body surface of a young dolphin usually has no spots or textures (for example, a young white dolphin), and in some scenarios, the image of the tail fin or dorsal fin of the young dolphin, which has biological identity characteristics, is not photographed in the obtained image of the young dolphin. Therefore, the identity information of the young dolphin includes the group mark.

[0107] In some specific examples, the identity information of the to-be-identified organism corresponding to the special group foreground image further includes the identity information of the group foreground image in the same group image. That is, the identity information of the young dolphin in the above example further includes the identity information of the adult dolphin appearing in the same group image at the same time.

[0108] In some specific examples, the biological identity identification method can further include the following steps: before performing foreground extraction on the obtained initial image of the to-be-identified organism, adaptively adjusting the initial image.

[0109] Specifically, the optical dimensions of the initial image such as brightness, contrast, color, and gray scale can be adaptively adjusted. For example, the brightness of the initial image can be adaptively adjusted by adaptive histogram equalization, automatic brightness correction, or contrast stretching. For example, adaptive white balance correction or color normalization processing can be used to adaptively adjust the color and white balance of the initial image. In addition, adaptive dynamic range compression technology can be used to set compression coefficients for the bright area, dark area, and middle tone area in the initial image, and unify the brightness dynamic range of the initial image to a target interval.

[0110] The optical indicators of the initial image can be uniformly adjusted to a preset range through adaptive adjustment, thereby improving the accuracy and comprehensiveness of feature extraction of the target object in the foreground region.

[0111] In some specific examples, the biological identity recognition method can further include the following steps: before calculating the similarity scores of the target foreground image and the plurality of sample images according to the preset similarity evaluation indicators, performing affine transformation operation on the target foreground image according to the sample images.

[0112] Specifically, the affine transformation operation can include translation, rotation, scaling, skewing and flipping, etc. The target foreground image and the sample image are matched and adjusted through the affine transformation operation, so that the target foreground image and the sample image as control groups are aligned in the global space.

[0113] In some specific examples, the biological identity recognition method can further include the following steps: determining the target control points in the target foreground image before the similarity calculation. In step S601, the image features of the target region can be extracted, which can be the features of the target control points in the target region, to obtain the target feature set.

[0114] Specifically, the target control points can be set according to the type of the to-be-identified biological, for example, when the to-be-identified biological is a cetacean, the target control points can be at least three feature points of the tip of the dorsal fin, the eye, the two bottom ends of the dorsal fin and the tip of the tail fin.

[0115] In some specific examples, the affine transformation operation on the target foreground image according to the sample image can include the following steps: after aligning the target foreground image and the corresponding control points in the sample image based on the target control points, performing the affine transformation operation.

[0116] In some specific examples, the biological identity recognition method can further include the following steps: before performing foreground extraction on the obtained initial image of the to-be-identified biological, preliminarily screening a plurality of original images of the to-be-identified biological to obtain the initial image of the to-be-identified biological.

[0117] Specifically, the preliminary screening of the original image can be performed according to a preset screening standard. The preset screening standard can include a quality standard and a feature standard. The quality standard can be determined by an image quality score, for example, the image quality score can be obtained according to the resolution, clarity, exposure, shooting angle and feature surface angle of the image. The feature standard can be the number of target objects included in the original image, for example, it can be required that the original image has at least a plurality of objects in a type of target objects (for example, a plurality of spots).

[0118] Regarding the screening standard for the preliminary screening of the original image, the screening standard can be set according to actual needs for different species of the to-be-identified biological in the original image.

[0119] Low quality images with insufficient information or unable to be identified can be removed through the preliminary screening.

[0120] It should be understood that, although Figures 2-7 The steps in the flowchart of the method of the first aspect are shown in sequence according to the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Figures 2-7 The steps shown and the steps involved in other embodiments are not strictly limited in order of execution unless explicitly stated herein. These steps can be executed in other orders. Moreover, at least a part of the steps of the foregoing embodiments can include multiple sub-steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in rotation or alternation with at least a part of other steps or sub-steps or stages of other steps.

[0121] In a second aspect, embodiments of the present disclosure provide a biological identity recognition system, as shown in Figure 8 The system can include a processor 810, a user terminal 820, and a sample database 830.

[0122] The processor 810 is configured to obtain an initial image of a biological subject to be identified from the user terminal 820, and execute the steps of the biological identity recognition method provided in any embodiment of the first aspect of the present disclosure.

[0123] The sample database 830 is configured to store a plurality of sample images, and the sample images are labeled with identity information of corresponding biological subjects.

[0124] In some embodiments, the processor 810 is further configured to compare the target foreground image with the plurality of sample images, calculate similarity scores of the target foreground image and the plurality of sample images, and determine a plurality of candidate sample images from the plurality of sample images according to the similarity scores, and send the plurality of candidate sample images and the target foreground image to the user terminal 820.

[0125] The user terminal 820 is configured to visually present the plurality of candidate sample images and the target foreground image, and feed back a confirmation result to the processor 810.

[0126] The processor 810 is further configured to determine a target sample image according to the confirmation result, and take identity information of the target sample image as target identity information of the target foreground image.

[0127] In a third aspect, embodiments of the present disclosure provide a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the biological identity recognition method provided in any of the embodiments of the first aspect of the present disclosure when executing the computer program.

[0128] In some embodiments, the computer device can be a server, and an internal structure diagram thereof can be as shown in Figure 9 The computer device comprises a processor, a memory, a network interface, and a database connected through a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The database of the computer device is configured to store sample images. The network interface of the computer device is configured to communicate with an external terminal through a network connection. The computer program is executed by the processor to implement the biological identity recognition method in any of the embodiments herein.

[0129] Those skilled in the art can understand that Figure 9 The structure shown in

[0130] In a fourth aspect, embodiments of the present disclosure provide a computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the steps of the biological identity recognition method provided in any of the embodiments of the first aspect of the present disclosure.

[0131] The computer readable storage medium can be Figure 9 The computer readable storage medium can be

[0132] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The aforementioned computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiment methods can be included. Any reference to memory, storage, database or other medium used in the embodiments of the present disclosure can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchnonous Link) DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0133] Any combination of the technical features of the above embodiments can be made, and in order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combination of the technical features does not exist contradictory, it should be considered as the range disclosed by the present disclosure.

[0134] The above embodiments only express some implementation manners of the present disclosure, and the description is more specific and detailed, but it should not be understood as a limitation on the protection scope of the present disclosure. It should be noted that for those skilled in the art, some modifications and improvements can be made without departing from the concept of the present disclosure, and these all belong to the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be subject to the appended claims.

Claims

1. A biometric identification method, characterized in that, The method includes: Foreground extraction is performed on the initial image of the organism to be identified to obtain a foreground image corresponding to the target region in the initial image. The target region includes target objects on the body surface of the organism to be identified. The target objects include at least one of the following: spots, textures, and parts with biological identification characteristics. The foreground image is subjected to feature enhancement processing to obtain the feature-enhanced target foreground image; The target foreground image is compared with multiple sample images to determine the target identity information of the organism to be identified corresponding to the target foreground image, wherein the sample images are marked with the identity information of the corresponding organism.

2. The method according to claim 1, characterized in that, The step of extracting the foreground from the initial image of the acquired organism to be identified, to obtain the foreground image corresponding to the target region in the initial image, includes: The initial image is segmented using a pre-defined deep learning network to obtain preliminary foreground information; The preliminary foreground information is visualized to obtain the foreground image.

3. The method according to claim 1, characterized in that, The step of comparing the target foreground image with multiple sample images to determine the target identity information of the organism to be identified corresponding to the target foreground image includes: Based on preset similarity evaluation indicators, the similarity score between the target foreground image and the multiple sample images is calculated. The similarity evaluation indicators include structural similarity index, two-dimensional correlation coefficient, mean square error value, and feature point matching distance. Based on the similarity score, a target sample image is determined from the plurality of sample images, wherein the target sample image and the target foreground image are images of the same organism at different times; The identity information of the target sample image is used as the target identity information of the target foreground image.

4. The method according to claim 3, characterized in that, The step of determining the target sample image from the plurality of sample images based on the similarity score includes: Based on the similarity score, multiple candidate sample images are determined from the multiple sample images; The multiple candidate sample images and the target foreground image are sent to the user terminal; The target sample image is determined based on the confirmation result fed back by the user terminal.

5. The method according to claim 3, characterized in that, The step of calculating the similarity score between the target foreground image and the multiple sample images based on a preset similarity evaluation index includes: Image features of the target region are extracted to obtain a target feature set, wherein the image features include the geometric attributes of the target object in the target region; Image features are extracted from the multiple sample images to obtain multiple sample feature sets corresponding to each sample image; The target feature set and the multiple sample feature sets are subjected to feature quantization processing to obtain the target feature vector and the multiple sample feature vectors; The target feature vector and the multiple sample feature vectors are calculated according to the preset similarity index to obtain multiple similarity scores corresponding to each sample image.

6. The method according to claim 1, characterized in that, The method further includes: Acquire a group image to be identified, wherein the group image includes multiple organisms to be identified; Foreground extraction is performed on the group images to obtain multiple group foreground images corresponding to the multiple organisms to be identified; Add group labels to the foreground images of the multiple groups, the group labels being used to indicate that there is a group relationship between the multiple organisms to be identified; When any of the group foreground images is a special group foreground image that does not contain the target object, the identity information of the organism to be identified corresponding to the special group foreground image includes the group tag.

7. A biometric identification system, characterized in that, The system includes a processor, a user terminal, and a sample database. The processor is configured to acquire an initial image of the organism to be identified from the user terminal and execute the steps of the method according to any one of claims 1 to 6; The sample database is used to store multiple sample images, each labeled with the identity information of the corresponding organism.

8. The system according to claim 7, characterized in that, The processor is further configured to: perform a similarity comparison between the target foreground image and the plurality of sample images; and calculate a similarity score between the target foreground image and the plurality of sample images; Based on the similarity score, multiple candidate sample images are determined from the multiple sample images; The plurality of candidate sample images and the target foreground image are sent to the user terminal; The user terminal is used to visualize the multiple candidate sample images and the target foreground image, and to feed back the confirmation result to the processor; The processor is further configured to determine the target sample image based on the confirmation result, and use the identity information of the target sample image as the target identity information of the target foreground image.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images