Target recognition method and apparatus, computer device, and readable storage medium

Abstract

Embodiments of the present application disclose a target recognition method and apparatus, a computer device, and a readable storage medium. The method comprises: acquiring an image to be processed, and performing face detection and key point detection on the image to be processed, so as to obtain a face image comprising a face region and coordinate information of key points in the face image; aligning the face image with a preset face standard image to obtain an aligned image; cropping, from the aligned image, a target image comprising a target key point region; and performing feature fusion on the aligned image and the target image to obtain fused data, and inputting the fused data into a network model to obtain a target recognition result. By applying the solution provided by an embodiment of the present application, feature fusion can be performed on the image comprising the face region and the target image comprising a key point region, and the fused data comprises more information of the key point region, so that the model can fully extract the features of the key point region, thereby improving the accuracy of target recognition.

Description

Target recognition method, device, computer equipment and readable storage medium Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a target recognition method, apparatus, computer equipment, and readable storage medium. Background Art

[0002] With the development of surveillance technology, target recognition is increasingly required in more and more scenarios, such as anti-fraud detection and facial key point detection in facial recognition applications. In some identity verification scenarios, anti-fraud detection requires verification of the user's identity. This technology effectively protects against attacks such as 2D static paper images, 2D static electronic images, 2D dynamic images, 3D masks, and 3D head models, as well as attacks using photos, videos, face swaps, masks, occlusions, 3D animations, and screen re-shoots, thereby identifying fraudulent activity and protecting user interests.

[0003] However, conventional techniques typically use the entire face image as input for target recognition after acquiring an image containing a face. However, since only a small portion of the ROI (region of interest) contributes significantly to recognition, and the ROI typically occupies a relatively small fraction of the total facial image pixel count, the model fails to fully extract information from the ROI, resulting in suboptimal recognition. Therefore, improving target recognition accuracy has become a pressing technical challenge. Summary of the Invention

[0004] This application provides a target recognition method, apparatus, computer device, and readable storage medium to improve the accuracy of target recognition. The specific technical solution is as follows.

[0005] In a first aspect, an embodiment of the present application provides a target recognition method, the method comprising:

[0006] Acquire an image to be processed, and perform face detection and key point detection on the image to be processed to obtain a facial image containing a facial area and coordinate information of each key point in the facial image;

[0007] Aligning the facial image with a preset facial standard image according to coordinate information of each key point in the facial image to obtain an aligned image;

[0008] intercepting a target image containing a target key point area from the aligned image;

[0009] The aligned image and the target image are subjected to feature fusion to obtain fused data, and the fused data is input into a network model to obtain a target recognition result.

[0010] In an embodiment of the present application, when performing target recognition, a target image containing a key point area can be obtained. The key point area is an area that plays an important role in target recognition. Then, feature fusion is performed on the image containing the facial area and the target image. The fused data will contain more information about the key point area, so that when performing target recognition, the model can fully extract the features of the key point area and improve the accuracy of target recognition.

[0011] Optionally, after intercepting the target image containing the target key point area in the aligned image, the method further includes:

[0012] Enlarging the target image to have the same width and height as the aligned image;

[0013] The step of performing feature fusion on the aligned image and the target image to obtain fused data comprises:

[0014] The aligned image and the target image are spliced ​​in the depth dimension, and the splicing result is used as the fused data.

[0015] Optionally, after intercepting the target image containing the target key point area in the aligned image, the method further includes:

[0016] Enlarging the target image to a height equal to that of the aligned image;

[0017] The step of performing feature fusion on the aligned image and the target image to obtain fused data comprises:

[0018] The aligned image and the target image are spliced ​​in the width dimension, and the splicing result is used as the fused data.

[0019] Optionally, the target key point area includes: an eye area; and the step of inputting the fused data into a network model to obtain a target recognition result includes:

[0020] The fused data is input into a network model to determine whether the human face included in the image to be processed is alive or non-living.

[0021] Optionally, the training process of the network model includes:

[0022] Obtain sample images and label categories corresponding to each sample image;

[0023] For each of the sample images, face detection and key point detection are performed on the sample image to obtain a sample face image containing a face area and coordinate information of each key point in the sample face image;

[0024] For each sample facial image, aligning the sample facial image with the preset facial standard image according to the coordinate information of each key point in the sample facial image to obtain an aligned sample image;

[0025] intercepting a target sample image containing a target key point area from each of the aligned sample images;

[0026] The aligned sample images and the corresponding target sample images are subjected to feature fusion to obtain fused data, and the fused data and the corresponding label categories are used for training to obtain the network model.

[0027] In a second aspect, an embodiment of the present application provides a target recognition device, the device comprising:

[0028] An image acquisition module is used to acquire an image to be processed, and perform face detection and key point detection on the image to be processed to obtain a facial image containing a facial area and coordinate information of each key point in the facial image;

[0029] an image processing module, configured to align the facial image with a preset facial standard image based on coordinate information of each key point in the facial image to obtain an aligned image;

[0030] An image capture module, configured to capture a target image containing a target key point region from the aligned image;

[0031] The target recognition module is used to perform feature fusion on the aligned image and the target image to obtain fused data, and input the fused data into the network model to obtain a target recognition result.

[0032] Optionally, the device further includes:

[0033] An image magnification module, configured to magnify the target image to have a width and height equal to those of the aligned image;

[0034] The target recognition module is specifically used to:

[0035] The aligned image and the target image are spliced ​​in the depth dimension, and the splicing result is used as the fused data.

[0036] Optionally, the device further includes:

[0037] An image magnification module, configured to magnify the target image to a height equal to that of the aligned image;

[0038] The target recognition module is specifically used to:

[0039] The aligned image and the target image are spliced ​​in the width dimension, and the splicing result is used as the fused data.

[0040] Optionally, the target key point area includes: an eye area; the target recognition module is specifically used to:

[0041] The fused data is input into a network model to determine whether the human face included in the image to be processed is alive or non-living.

[0042] Optionally, the device further includes a model training module, specifically configured to:

[0043] Obtain sample images and label categories corresponding to each sample image;

[0044] For each of the sample images, face detection and key point detection are performed on the sample image to obtain a sample face image containing a face area and coordinate information of each key point in the sample face image;

[0045] For each sample facial image, aligning the sample facial image with the preset facial standard image according to the coordinate information of each key point in the sample facial image to obtain an aligned sample image;

[0046] intercepting a target sample image containing a target key point area from each of the aligned sample images;

[0047] The aligned sample images and the corresponding target sample images are subjected to feature fusion to obtain fused data, and the fused data and the corresponding label categories are used for training to obtain the network model.

[0048] In a third aspect, an embodiment of the present application provides a computer device, comprising: a memory and a processor, wherein the memory and the processor are coupled;

[0049] The memory is used to store one or more computer instructions;

[0050] The processor is used to execute the one or more computer instructions to implement the target recognition method as described in the first aspect.

[0051] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having one or more computer instructions stored thereon, which are executed by a processor to implement the target recognition method as described in the first aspect above.

[0052] In a fifth aspect, the present application provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, it implements the target recognition method described in the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] To more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art. Obviously, the drawings described below are merely some embodiments of the present application. Those skilled in the art can derive other drawings based on these drawings without inventive effort.

[0054] FIG1 is a schematic diagram showing a comparison of infrared images of living and non-living eyes;

[0055] FIG2 is a schematic diagram of a live body recognition network model training process in an embodiment of the present application;

[0056] FIG3 is a flow chart of a target recognition method provided in an embodiment of the present application;

[0057] FIG4 is a schematic diagram of a subsequent image and an amplified target image according to an embodiment of the present application;

[0058] FIG5 is a schematic structural diagram of a target recognition device provided in an embodiment of the present application;

[0059] FIG6 is a schematic diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0060] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0061] It should be noted that the terms "including" and "having" and any variations thereof in the embodiments and drawings of this application are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or device comprising a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to the process, method, product, or device.

[0062] The embodiments of the present application disclose a target recognition method, apparatus, computer device, and readable storage medium, which can improve the accuracy of target recognition. The embodiments of the present application are described in detail below.

[0063] In some identity verification scenarios for anti-fraud detection, it's necessary to determine whether a user is truly alive. This means the model must identify a live or non-live person based on facial images captured by surveillance equipment. However, the applicant's research found significant differences in the eye features between live and non-live individuals. This means that the five facial features are not equally important in liveness detection, with the eyes playing a key role.

[0064] Please refer to Figure 1, which shows a comparison of infrared images of living and non-living eyes. The two images on the left are live images, while the images on the right, from top to bottom, show the corresponding paper mask and 3D-simulated mask images for the live images on the left. As can be seen from Figure 1, the eye features of living and non-living subjects, corresponding to different attack methods, such as the paper mask and 3D-simulated mask, differ significantly. Especially with mask attacks, the differences between living and non-living eyes are significant. Generally, non-living eye features lack the high-frequency information of living eyes and are less detailed. For example, the paper mask is completely empty.

[0065] In facial landmark detection, the importance of facial features is also not equal. The key points detected are the areas that require the most attention in image recognition. For example, when detecting the eyes, the eye area is the area of ​​interest; when detecting the mouth, the mouth area is the area of ​​interest.

[0066] It can be seen that when performing target recognition such as liveness recognition or key point detection, there are ROI areas that need to be paid attention to. Based on this, in an embodiment of the present application, a target recognition method is proposed, which can extract more image features of the ROI area during the recognition process, thereby improving the accuracy of target recognition. For example, when performing liveness recognition, the eye area can be selected as a specific ROI area, and the local feature differences of the ROI area can be fused and input, so that the network model can better learn the different features of living and non-living objects, thereby improving the accuracy of liveness recognition.

[0067] In the embodiment of the present application, liveness recognition can be taken as an example to illustrate the target recognition method provided in the embodiment of the present application. The processes of other key point detection differ only in the selection of key points. The embodiment of the present application does not provide detailed descriptions of other processes.

[0068] Before performing liveness recognition, the network model can be trained first, for example, the network model can be trained offline. Specifically, sample images and the label categories corresponding to each sample image can be obtained; for each sample image, face detection and key point detection are performed on the sample image to obtain a sample face image containing a face area and the coordinate information of each key point in the sample face image; for each sample face image, the sample face image is aligned with a preset face standard image according to the coordinate information of each key point in the sample face image to obtain an aligned sample image; a target sample image containing a target key point area is intercepted from each aligned sample image; each aligned sample image and the corresponding target sample image are subjected to feature fusion to obtain fused data, and each fused data and the corresponding label category are used to train a network model.

[0069] When training a network model for liveness recognition, the sample images may include live body images, as well as various types of non-live body images, such as two-dimensional static paper images, two-dimensional static electronic images, two-dimensional dynamic images, three-dimensional masks, and three-dimensional head models. The target key point area may include the eye area.

[0070] As shown in Figure 2, it shows a schematic diagram of the liveness recognition network model training process in an embodiment of the present application. First, face detection and key point detection can be performed on the acquired original image to obtain the face detection frame and the coordinates of the facial key points. Then, the face detection frame is aligned with the five points of the standard face image. Then, the eye area is cropped out using the eye key points. After the eye area is enlarged to the size of the face frame, the entire face and the enlarged eye features are fused as the network input, and then the liveness recognition model is trained according to the label category.

[0071] FIG3 is a flow chart of a target recognition method provided in an embodiment of the present application. The method can be applied to electronic devices and includes the following steps:

[0072] S310: Acquire an image to be processed, and perform face detection and key point detection on the image to be processed to obtain a facial image including a facial area and coordinate information of each key point in the facial image.

[0073] The above-mentioned images to be processed may be images that require target recognition, for example, images collected in real time by monitoring equipment, pre-stored images, etc.

[0074] It's understood that the image to be processed typically includes background areas in addition to the facial region. However, during target recognition, only the facial region needs to be identified. Therefore, after acquiring the image to be processed, face detection and key point detection can be performed on the image. Key points can include facial features. This can yield a facial image containing the facial region and the coordinates of each key point within the facial image. In other words, an image containing only the facial region can be cropped from the image to be processed to serve as the facial image.

[0075] The size of the facial image can be pre-set to a fixed size, or can be determined based on the size of the facial region in the image to be processed. The embodiment of this application does not impose any specific limitation on the size of the facial image. In other words, the electronic device can extract an image of a fixed size that contains the facial region from the image to be processed as the facial image, or it can extract a minimum image that contains only the facial region from the image to be processed as the facial image.

[0076] Furthermore, an image to be processed may include one or more faces, and thus, the facial images obtained from an image to be processed may be one or more. When multiple facial images are obtained, the electronic device may sequentially perform target recognition on each facial image according to subsequent steps to obtain multiple recognition results.

[0077] S320: Aligning the facial image with a preset facial standard image based on the coordinate information of each key point in the facial image to obtain an aligned image.

[0078] It is understood that the face portion of the image to be processed may be tilted, distorted, etc., which will affect the accuracy of target recognition. In the embodiment of the present application, the facial image can be aligned based on a pre-stored facial standard image to ensure the accuracy of target recognition.

[0079] For example, based on the coordinate information of the key points of the facial features in the facial image, the facial features in the facial image can be adjusted in sequence to match the preset facial standard image, and the adjusted image is used as the aligned image.

[0080] S330: intercepting a target image including a target key point area from the aligned image.

[0081] The target key point area can be set according to specific needs. For example, in living body recognition, the target key point area can be set to the eye area; in key point detection, the target key point area is the corresponding key point area.

[0082] Specifically, the electronic device can capture the smallest image containing only the target key point area in the aligned image as the target image; or, it can capture an image containing the target key point area and of a preset fixed size as the target image, both of which are possible.

[0083] S340: Perform feature fusion on the aligned image and the target image to obtain fused data, and input the fused data into the network model to obtain the target recognition result.

[0084] After obtaining the target image, the electronic device can perform feature fusion on the aligned image and the target image to generate fused data. Feature fusion can produce richer feature representations, and the fused data contains more feature information about key point areas. Using this fused data as input to the network model can improve target recognition accuracy.

[0085] For example, the electronic device may use a depth dimension fusion method to perform feature fusion on the aligned image and the target image to obtain fused data; or, it may use a width dimension fusion method to perform feature fusion on the aligned image and the target image to obtain fused data.

[0086] The stitching requirements of the depth dimension can be different in the depth dimension of the connection, but must be equal in the height and width dimensions. Therefore, when the depth dimension stitching method is used to fuse the aligned image and the target image, before fusion, the target image needs to be enlarged to be equal to the width and height of the aligned image; then the aligned image and the target image can be stitched in the depth dimension, and the stitching result can be used as the fused data.

[0087] For example, when the aligned image is batch×1×224×224, the target image is enlarged and concatenated with the aligned image in the depth dimension to batch*2*224*224. The input of the model is batch×2×224×224.

[0088] When enlarging the target image, for example, the target image can first be stretched to the same width as the aligned image, and then filled with black in the height direction to obtain a target image with the same width and height as the aligned image. As shown in Figure 4, the aligned image and the enlarged target image are shown from left to right. As can be seen from Figure 4, compared to the aligned image, that is, the image including the entire face, the enlarged target image has more pixels in the eye area. Therefore, when performing liveness recognition, the feature information of the eye area can be better extracted, improving the accuracy of liveness recognition.

[0089] The stitching requirements of the width dimension can be different in the width dimension of the connection, but must be equal in the height and depth dimensions. Therefore, when the width dimension stitching method is used to fuse the aligned image and the target image, before fusion, the target image needs to be enlarged to the same height as the aligned image; then the aligned image and the target image can be stitched in the width dimension, and the stitching result can be used as the fused data.

[0090] For example, when the aligned image is batch×1×224×224, the target image is enlarged and concatenated with the aligned image in the width dimension to batch*1*224*448. The input of the model is batch×1×224×448.

[0091] In an embodiment of the present application, when performing target recognition, a target image containing a key point area can be obtained. The key point area is an area that plays an important role in target recognition. Then, feature fusion is performed on the image containing the facial area and the target image. The fused data will contain more information about the key point area, so that when performing target recognition, the model can fully extract the information about the key point area and improve the accuracy of target recognition.

[0092] When performing liveness recognition, the eye area is selected as a specific ROI area, which is enlarged to the same size as the face and its features are extracted. This area is then fused with the features extracted from the overall facial image and input into the model for classification. This method takes advantage of the obvious information difference between the eyes of living and non-living objects. By fusion input, the network can learn better, improve the model's expressive ability, and enhance the recognition accuracy of living and non-living objects.

[0093] When evaluating liveness recognition, the values ​​of TP (True Positive, predicted to be live, actually live), TN (True Negative, predicted to be non-live, actually non-live), FP (False Positive, predicted to be live, actually non-live), and FN (False Negative, predicted to be non-live, actually live) can be calculated based on the results of the test dataset, and the values ​​of TPR (True Positive Rate) and FPR (False Positive Rate) can be calculated based on the following formulas (1) and (2). TPR is called the true positive rate, also known as the hit rate or sensitivity. Similar to the recall rate or recall rate, it represents the proportion of actual positive instances predicted correctly to all positive instances. FPR is called the false positive rate, also known as the false hit rate or false alarm rate, which represents the proportion of actual negative instances predicted correctly to all negative instances. By drawing a PR curve based on the calculated TPR and FPR, the corresponding TPR value under different FPR values ​​can be obtained. TPR=TP / (TP+FN) (1) FPR=FP / (FP+TN) (2)

[0094] The model training was performed using facial features alone, facial and eye features depth fusion as input, and facial and eye features width fusion as input. The comparison training results are shown in the following table:

[0095] The table above compares the three experimental groups. Using a deep fusion of facial and eye features as input, the model achieved the highest recognition accuracy, with a TPR of 96.5% at an FPR of 0.01. Using facial features alone as input, the model achieved the lowest recognition accuracy, with a TPR of 90.5% at an FPR of 0.01. This demonstrates that the liveness detection input design based on the fusion of facial and ROI features can leverage information differences within the ROI region, enabling the network model to better learn the distinct characteristics of live and non-live subjects and improving recognition accuracy.

[0096] FIG5 shows a schematic diagram of the structure of a target recognition device provided by an embodiment of the present application, wherein the device includes:

[0097] An image acquisition module 510 is configured to acquire an image to be processed and perform face detection and key point detection on the image to be processed to obtain a facial image containing a facial region and coordinate information of each key point in the facial image;

[0098] An image processing module 520 is configured to align the facial image with a preset facial standard image based on the coordinate information of each key point in the facial image to obtain an aligned image;

[0099] An image capture module 530 is configured to capture a target image containing a target key point region from the aligned image;

[0100] The target recognition module 540 is used to perform feature fusion on the aligned image and the target image to obtain fused data, and input the fused data into the network model to obtain a target recognition result.

[0101] Optionally, the device further includes:

[0102] An image magnification module, configured to magnify the target image to have a width and height equal to those of the aligned image;

[0103] The target identification module 540 is specifically configured to:

[0104] The aligned image and the target image are spliced ​​in the depth dimension, and the splicing result is used as the fused data.

[0105] Optionally, the device further includes:

[0106] An image magnification module, configured to magnify the target image to a height equal to that of the aligned image;

[0107] The target identification module 540 is specifically configured to:

[0108] The aligned image and the target image are spliced ​​in the width dimension, and the splicing result is used as the fused data.

[0109] Optionally, the target key point area includes: an eye area; the target recognition module 540 is specifically configured to:

[0110] The fused data is input into a network model to determine whether the human face included in the image to be processed is alive or non-living.

[0111] Optionally, the device further includes a model training module, specifically configured to:

[0112] Obtain sample images and label categories corresponding to each sample image;

[0113] For each of the sample images, face detection and key point detection are performed on the sample image to obtain a sample face image containing a face area and coordinate information of each key point in the sample face image;

[0114] For each sample facial image, aligning the sample facial image with the preset facial standard image according to the coordinate information of each key point in the sample facial image to obtain an aligned sample image;

[0115] intercepting a target sample image containing a target key point area from each of the aligned sample images;

[0116] The aligned sample images and the corresponding target sample images are subjected to feature fusion to obtain fused data, and the fused data and the corresponding label categories are used for training to obtain the network model.

[0117] In an embodiment of the present application, when performing target recognition, a target image containing a key point area can be obtained. The key point area is an area that plays an important role in target recognition. Then, feature fusion is performed on the image containing the facial area and the target image. The fused data will contain more information about the key point area, so that when performing target recognition, the model can fully extract the information about the key point area and improve the accuracy of target recognition.

[0118] The above device embodiment corresponds to the method embodiment and has the same technical effects as the method embodiment. For detailed description, please refer to the method embodiment. The device embodiment is obtained based on the method embodiment. For detailed description, please refer to the method embodiment part and will not be repeated here.

[0119] Next, a computer device provided in an embodiment of the present application is described. Please refer to FIG6 , which is a schematic diagram of the structure of a computer device provided in an embodiment of the present application. The computer device includes:

[0120] one or more processors 40;

[0121] The processor 40 is coupled to a storage device 41, and the storage device 41 is used to store one or more programs.

[0122] When the one or more programs are executed by the one or more processors 40 , the electronic device implements a technical solution of a target recognition method as described in FIG. 1 to FIG. 4 .

[0123] An embodiment of the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a technical solution of a target recognition method as described in FIG. 1 to FIG. 4 .

[0124] The present application provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the technical solution of a target recognition method as described in Figures 1 to 4.

[0125] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of an embodiment, and the modules or processes in the accompanying drawings are not necessarily required to implement the present application.

[0126] Those skilled in the art will appreciate that the modules in the apparatuses of the embodiments may be distributed in the apparatuses of the embodiments as described in the embodiments, or may be located in one or more apparatuses different from the embodiments with corresponding changes. The modules in the above embodiments may be combined into one module or further divided into multiple sub-modules.

[0127] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A target recognition method, characterized in that: The method comprises: Acquire an image to be processed, and perform face detection and key point detection on the image to be processed to obtain a facial image containing a facial area and coordinate information of each key point in the facial image; Aligning the facial image with a preset facial standard image according to coordinate information of each key point in the facial image to obtain an aligned image; intercepting a target image containing a target key point area from the aligned image; The aligned image and the target image are subjected to feature fusion to obtain fused data, and the fused data is input into a network model to obtain a target recognition result.

2. The method according to claim 1, characterized in that After intercepting the target image containing the target key point area in the aligned image, the method further includes: Enlarging the target image to have the same width and height as the aligned image; The step of performing feature fusion on the aligned image and the target image to obtain fused data comprises: The aligned image and the target image are spliced in the depth dimension, and the splicing result is used as the fused data.

3. The method according to claim 1, characterized in that After intercepting the target image containing the target key point area in the aligned image, the method further includes: Enlarging the target image to a height equal to that of the aligned image; The step of performing feature fusion on the aligned image and the target image to obtain fused data comprises: The aligned image and the target image are spliced in the width dimension, and the splicing result is used as the fused data.

4. The method according to claim 1, wherein The target key point area includes: the eye area; the step of inputting the fused data into the network model to obtain the target recognition result includes: The fused data is input into a network model to determine whether the human face included in the image to be processed is alive or non-living.

5. The method according to any one of claims 1 to 4, characterized in that The training process of the network model includes: Obtain sample images and label categories corresponding to each sample image; For each of the sample images, face detection and key point detection are performed on the sample image to obtain a sample face image containing a face area and coordinate information of each key point in the sample face image; For each sample facial image, aligning the sample facial image with the preset facial standard image according to the coordinate information of each key point in the sample facial image to obtain an aligned sample image; intercepting a target sample image containing a target key point area from each of the aligned sample images; The aligned sample images and the corresponding target sample images are subjected to feature fusion to obtain fused data, and the fused data and the corresponding label categories are used for training to obtain the network model.

6. A target recognition device, characterized in that: The device comprises: An image acquisition module is used to acquire an image to be processed, and perform face detection and key point detection on the image to be processed to obtain a facial image containing a facial area and coordinate information of each key point in the facial image; an image processing module, configured to align the facial image with a preset facial standard image based on coordinate information of each key point in the facial image to obtain an aligned image; An image capture module, configured to capture a target image containing a target key point region from the aligned image; The target recognition module is used to perform feature fusion on the aligned image and the target image to obtain fused data, and input the fused data into the network model to obtain a target recognition result.

7. The device according to claim 6, characterized in that The device further comprises: An image magnification module, configured to magnify the target image to have a width and height equal to those of the aligned image; The target recognition module is specifically used to: splice the aligned image and the target image in the depth dimension, and use the splicing result as fused data.

8. The device according to claim 6, characterized in that The device further comprises: An image magnification module, configured to magnify the target image to a height equal to that of the aligned image; The target recognition module is specifically configured to: splice the aligned image and the target image in a width dimension, and use the splicing result as fused data.

9. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are coupled; The memory is used to store one or more computer instructions; The processor is configured to execute the one or more computer instructions to implement the target recognition method according to any one of claims 1 to 5.

10. A readable storage medium having one or more computer instructions stored thereon, characterized in that: The instruction is executed by a processor to implement the target recognition method according to any one of claims 1 to 5.

Images