A biometric identification method and apparatus
By using multimodal image segmentation and hierarchical fusion model recognition for biometric features, the problems of low accuracy and slow speed in biometric recognition are solved, achieving efficient and accurate biometric recognition under the influence of environmental factors.
Patent Information
- Application Number
- CN202210049340.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-01-17
- Publication Date
- 2026-01-27
- Estimated Expiration
- 2042-01-17
AI Technical Summary
Existing biometric identification methods suffer from low accuracy and slow speed. In particular, under the influence of factors such as lighting and biometric acquisition angle, single image recognition is fast but inaccurate, while block processing improves accuracy but is slow.
Images of multiple different modalities of the target biometric feature are segmented into image block groups, and a multi-layer fusion model is called for recognition. The layering has a preset order, and recognition stops when the recognition result of any layer meets the conditions. The biometric feature recognition result is determined based on the current result.
While improving the accuracy of biometric identification, it reduces identification time, increases identification speed, and reduces the impact of environmental factors.
Smart Images
Figure CN114550314B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of biometric identification technology, and in particular to a biometric identification method and apparatus. Background Technology
[0002] Human biometrics are widely used in the field of identity recognition due to their uniqueness, privacy, and immutability.
[0003] Currently, the two common biometric identification methods for human biometrics are as follows: First, using a biometric identification model to identify biometric features from a single image. While this method is relatively fast, image quality can be poor due to factors such as lighting and the angle at which the image is captured, resulting in lower accuracy. Second, dividing a single image into blocks and then using a biometric identification model to identify biometric features from each block. Although this method offers improved accuracy compared to the first, the large number of image blocks significantly slows down the identification process, resulting in lower overall speed.
[0004] It is evident that existing biometric identification methods suffer from low accuracy and slow speed. Summary of the Invention
[0005] In view of this, the present invention proposes a biometric identification method and device, the main purpose of which is to improve the accuracy of biometric identification while increasing the identification speed.
[0006] In a first aspect, the present invention provides a biometric identification method, the method comprising:
[0007] At least two images of the target biometric feature are segmented to form image block groups for each image, wherein each image has a different modality;
[0008] A layered fusion model composed of multiple layers is invoked to identify each of the image block groups. Each layer is used to identify one image block group, and each layer has a preset identification order. After any layer identifies the corresponding image block group and outputs the layer identification result, the layer immediately following it in the sequence begins to identify the corresponding image block group.
[0009] When the conditions for stopping the recognition process of the hierarchical fusion model are met, the biometric recognition result is determined based on the currently obtained hierarchical recognition results.
[0010] In a second aspect, the present invention provides a biometric identification device, the device comprising:
[0011] A segmentation unit is used to segment at least two images of a target biometric feature into blocks, forming image block groups for each image, wherein each image has a different modality;
[0012] The calling unit is used to call a layered fusion model composed of multiple layers to perform recognition processing on each of the image block groups. Here, one layer is used to identify one image block group, and each layer has a preset recognition order. After any layer identifies the corresponding image block group and outputs the layer recognition result, the layer immediately following it in the sequence begins to identify the corresponding image block group.
[0013] The determining unit is used to determine the biometric recognition result based on the currently obtained hierarchical recognition result when it is determined that the condition for stopping the recognition process of the hierarchical fusion model is met.
[0014] Thirdly, the present invention provides a computer-readable storage medium comprising a stored program, wherein, when the program is executed, it controls the device on which the storage medium is located to perform the biometric identification method described in the first aspect.
[0015] Fourthly, the present invention provides an electronic device, the electronic device comprising:
[0016] Memory, used to store programs;
[0017] A processor, coupled to the memory, is used to run the program to perform the biometric identification method described in the first aspect.
[0018] By employing the above technical solution, the biometric recognition method and apparatus provided by this invention, when biometric recognition is required, divides multiple images of different modalities of the target biometric feature into image block groups. Then, a hierarchical fusion model composed of multiple layers is invoked to perform recognition processing on each image block group. One layer in the hierarchical fusion model is used to recognize one image block group, and each layer has a preset recognition order. When the hierarchical fusion model performs biometric recognition, after any layer recognizes its corresponding image block group and outputs its hierarchical recognition result, the layer immediately following it begins recognizing its corresponding image block group. When the conditions for stopping the hierarchical fusion model's recognition processing are met, the biometric recognition result is determined based on the currently obtained hierarchical recognition results. Therefore, this invention, in order to reduce the impact of environmental factors such as lighting and biometric acquisition angle on the accuracy of biometric recognition, performs biometric recognition on image block groups after dividing multiple images of different modalities of the target biometric feature into blocks. Furthermore, to improve the speed of biometric recognition, once all the currently obtained hierarchical recognition results meet the requirements for generating the target biometric recognition result, the hierarchical fusion model can stop biometric recognition. This eliminates the need for the hierarchical fusion model to complete the recognition of all image blocks; the biometric recognition result can be determined based on the currently obtained hierarchical recognition results, thereby reducing the biometric recognition time and accelerating the recognition process. In summary, the solution provided by this invention can improve both the accuracy and speed of biometric recognition.
[0019] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description
[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0021] Figure 1 A flowchart of a biometric identification method provided in one embodiment of the present invention is shown;
[0022] Figure 2 A flowchart of a biometric identification method provided by another embodiment of the present invention is shown;
[0023] Figure 3 This diagram illustrates a multispectral palmprint acquisition device according to an embodiment of the present invention.
[0024] Figure 4 This diagram illustrates a structural schematic of a biometric identification device according to an embodiment of the present invention.
[0025] Figure 5 A schematic diagram of a biometric identification device provided in another embodiment of the present invention is shown. Detailed Implementation
[0026] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
[0027] Humans possess biometric features such as fingerprints, palm prints, and irises. These biometric features are inherent physiological characteristics of the human body, possessing properties such as uniqueness, privacy, and immutability, and are therefore widely used in the field of identity verification.
[0028] Currently, the two common biometric identification methods for human biometrics are as follows: First, using a biometric identification model to identify biometric features from a single image. While this method is relatively fast, image quality can be poor due to factors such as lighting, acquisition angle, and rotation, resulting in lower accuracy. Second, dividing a single image into blocks and then using a biometric identification model to identify biometric features from each block. Although this method offers improved accuracy compared to the first, the large number of image blocks significantly impacts the speed of biometric identification, leading to a lower overall speed.
[0029] It is evident that existing biometric identification methods suffer from low accuracy and slow speed. To overcome these shortcomings, this invention proposes a biometric identification method and apparatus to improve both accuracy and speed. The specific application scenarios of the proposed biometric identification method and apparatus can be determined based on business requirements, and are not specifically limited in this embodiment. For example, the proposed biometric identification method and apparatus can be applied to identity verification scenarios such as banking and military applications. The following provides a detailed description of the proposed biometric identification method and apparatus.
[0030] like Figure 1 As shown, this embodiment of the invention provides a biometric identification method, which mainly includes:
[0031] 101. At least two images of the target biometric feature are segmented to form image block groups for each image, wherein each image has a different modality.
[0032] In practical applications, biometric identification is a prerequisite for identity verification. Therefore, the target biometric is the biological characteristic possessed by the human body to be identified. The selection of the target biometric depends on the specific identity verification scenario, and this embodiment does not impose specific limitations. Optionally, the target biometric can be any one of the following: fingerprint, palm print, and iris scan.
[0033] For example, the biometric identification method provided in this embodiment of the invention targets palm prints. The advantages of choosing palm prints are as follows: First, the size of the palm is larger than that of the fingers and eyes, therefore the acquired palm print image can include richer features. Second, palm print acquisition is more secure than fingerprint acquisition. The fingertip where fingerprints are located is raised, so fingerprint acquisition requires pressing the fingertip against the fingerprint acquisition surface of the fingerprint acquisition device, which can leave fingerprint residue and be maliciously stolen. The palm surface where palm prints are located is nearly flat, so palm print acquisition can be done without contact, thus avoiding the possibility of palm print residue on the acquisition device being stolen by malicious individuals. Third, palm print acquisition is more convenient than iris scanning. Iris scanning requires the user to align their eyes with the iris acquisition device, which restricts the user's head position and thus limits their movement, resulting in a poor user experience. Palm print acquisition only requires the user to extend their palm, with less restriction on movement and a relatively better user experience. Fourth, palm prints have a unique characteristic. Even among people with identical or similar facial features, palm prints are unique, making them more secure.
[0034] For example, in the identity recognition scenario of an access control system, palm prints are used as the target biometric feature required for identity recognition in access control systems due to their rich features and ease of acquisition.
[0035] After identifying the target biometric features, images of the target features of the human body to be identified need to be acquired. To reduce the impact of environmental factors such as lighting and biometric acquisition angle on the accuracy of biometric recognition, two or more images of the target biometric features are acquired, and each image has a different modality. The modality defines the acquisition scene of the images; that is, each image is acquired in a different scene.
[0036] For example, the target biometric feature is a palm print, and three images of the palm print were acquired, namely Image 1, Image 2, and Image 3. Image 1 corresponds to the blue spectral mode, meaning it was acquired under the blue spectrum. Image 2 corresponds to the green spectral mode, meaning it was acquired under the green spectrum. Image 3 corresponds to the red spectral mode, meaning it was acquired under the red spectrum.
[0037] After acquiring images of the target biometric features, the images need to be segmented into blocks to form separate image block groups. The image segmentation method is described below:
[0038] For each image, the following steps are performed: Set the number of image blocks for the image; determine the block size based on the number of image blocks and the image size; and divide the image into blocks based on the block size. For example, if the number of image blocks for image 1 is set to 64 and the size of image 1 is 16×16, then image 1 will be divided into 64 image blocks of size 2×2, and the image block group of image 1 will contain 64 image blocks of size 2×2.
[0039] It should be noted that all images of the target biometric feature are the same size. The number and size of image blocks in the image block groups of each image of the target biometric feature can exist in the following ways: First, the number of image blocks in each image block group is different. For example, there are images 1, 2, and 3. The image block group of image 1 includes 4 image blocks of size 8×8, the image block group of image 2 includes 16 image blocks of size 4×4, and the image block group of image 3 includes 64 image blocks of size 2×2. Second, the number of image blocks in each image block group is the same. For example, there are images 1, 2, and 3. The image block groups of images 1, 2, and 3 all include 64 image blocks of size 2×2. The third type involves different numbers of image blocks in the image block groups of each image. For example, there are images 1, 2, and 3. The image block group of image 1 includes 4 image blocks of size 8×8, while the image block groups of images 2 and 3 each include 64 image blocks of size 2×2.
[0040] 102. Call the layered fusion model composed of multiple layers to perform recognition processing on each of the image block groups. Here, one layer is used to identify one image block group, and each layer has a preset recognition order. After any layer identifies the corresponding image block group and outputs the layer recognition result, the layer immediately following it in the sequence begins to identify the corresponding image block group.
[0041] The layered fusion model consists of multiple layers used for biometric identification of images. Each layer in the model has a preset identification order, with one layer corresponding to one image block group. Within this preset order, after any layer identifies its corresponding image block group and outputs its identification result, the next layer in the sequence begins identifying its corresponding image block group. The purpose of this layered identification method, following a preset order, is that once the identification result of a certain layer satisfies the requirements for generating the target biometric identification result, the layered fusion model can stop its biometric identification process. Based on the currently obtained layered identification results, the biometric identification result can be determined, thereby reducing the identification time and accelerating the biometric identification process.
[0042] It should be noted that the type of layered fusion model can be determined based on specific business needs, and this embodiment does not impose any specific limitations. Optionally, the layered fusion model can be a layered transform model or a multilayer perceptron (MLP) model.
[0043] The following describes the specific process of using a hierarchical fusion model composed of multiple layers to identify and process each image block group. This process includes the following steps one and two:
[0044] Step 1: Determine the image block groups corresponding to each layer in the layered fusion model.
[0045] In practical applications, the process of determining the image block groups corresponding to each layer in the hierarchical fusion model is related to the number of image blocks in each image block group. Therefore, the determination process can be divided into the following situations:
[0046] The first approach is to determine the image block group corresponding to each layer when the number of image blocks in each image block group is different. In this approach, the layer that is ranked earlier in the preset recognition order of each layer will have fewer image blocks in its corresponding image block group.
[0047] Because layers listed earlier in the preset recognition order are identified first, during the layer-by-layer recognition process, once the layer recognition result of a certain layer is output, and all the currently obtained layer recognition results meet the requirements for generating the target biometric recognition result, the biometric recognition of the layers following it can be stopped. Based on the currently obtained layer recognition results, the biometric recognition result is determined. This reduces the recognition time for biometrics and speeds up the recognition process.
[0048] For example, the layered fusion model consists of layer 1, layer 2, and layer 3. The preset recognition order for each layer is: layer 1, layer 2, and layer 3. Given images 1, 2, and 3, the order of the number of image blocks in each image's block group from smallest to largest is: image 1, image 2, and image 3. Therefore, layer 1 corresponds to the image block group of image 1, layer 2 corresponds to the image block group of image 2, and layer 3 corresponds to the image block group of image 3.
[0049] The second method, when the number of image blocks in each image block group is the same, involves determining the image block group corresponding to each layer in two ways: One is to randomly determine the image block group corresponding to each layer, ensuring that the correspondence between each layer and the image block group is unique. The other is to determine the image block group of each layer based on the modality of the image to which the image block group belongs. For example, if there is a set modality recognition order, then the corresponding image block group for each layer is determined based on the modality recognition order. For example, consider images 1, 2, and 3. Image 1 corresponds to the blue spectral modality, image 2 corresponds to the green spectral modality, and image 3 corresponds to the red spectral modality. The modality recognition order from front to back is: blue spectral modality, green spectral modality, red spectral modality. The preset recognition order for each layer is: layer 1, layer 2, and layer 3. Then, the image block group corresponding to image 1 is determined as layer 1, the image block group corresponding to image 2 is determined as layer 2, and the image block group corresponding to image 3 is determined as layer 3.
[0050] The third scenario involves image groups containing different numbers of image blocks. For example, consider images 1, 2, and 3. Image 1's image group contains 4 image blocks, while images 2 and 3 each contain 64 image blocks. The preset recognition order is: Layer 1, Layer 2, and Layer 3. The order of the number of image blocks in each image group, from fewest to most, is: Image 1, Image 2, and Image 3. In this order, Image 2 and Image 3 have the same number of blocks, so Image 2 is randomly placed before Image 3. Therefore, Layer 1 corresponds to the image group containing Image 1, Layer 2 corresponds to the image group containing Image 2, and Layer 3 corresponds to the image group containing Image 3.
[0051] Step 2: Sequentially call the corresponding image block groups for each layer to identify their respective segments.
[0052] After determining the image block groups corresponding to each layer, the corresponding image block groups of each layer are identified in sequence. It should be emphasized that when each layer is identified, the image block groups of the currently used layer are identified and the layer identification results are output. Then, the layer immediately following it will start to identify the corresponding image block groups.
[0053] 103. When it is determined that the conditions for stopping the recognition process of the hierarchical fusion model are met, the biometric recognition result is determined based on the currently obtained hierarchical recognition result.
[0054] To promptly determine the results of biometric identification, it is necessary to ascertain whether the conditions for the hierarchical fusion model to cease identification processing are met. If these conditions are met, the hierarchical fusion model will be stopped from performing biometric identification. The process of determining whether the conditions for the hierarchical fusion model to cease identification processing includes the following steps one through five:
[0055] Step 1: Determine the latest obtained hierarchical identification result.
[0056] The latest layer recognition results are the latest layer recognition results output by the layer that has undergone biometric recognition.
[0057] The hierarchical recognition result includes a recognition score and a target object. The recognition score reflects the probability that the image corresponding to the hierarchical recognition result is an image with the biometric features of the target object. For example, the hierarchical recognition result of layer 1 for image 1 includes a recognition score of "80%" and a target object "Zhang San", indicating that the probability that image 1 is an image with the biometric features of Zhang San is 80%.
[0058] Step 2: Determine whether the recognition score of the latest layered recognition result has reached the first threshold. If not, proceed to Step 3; otherwise, proceed to Step 4.
[0059] The hierarchical fusion model consists of multiple layers. Theoretically, the sum of the recognition scores in the hierarchical recognition results output by each layer should be a predetermined value, such as 100%. Therefore, a first threshold needs to be set. If the recognition score in the hierarchical recognition result of a layer reaches the first threshold, it indicates that the accuracy of that layer's recognition result is high. Even if a layer in the preset recognition order performs biometric recognition, the accuracy of its resulting hierarchical recognition result will not be higher than the accuracy of its output. Therefore, to promptly determine the biometric recognition result, it is necessary to check whether the recognition score included in the latest obtained hierarchical recognition result reaches the first threshold.
[0060] If the latest layered identification result is determined to have an identification score that reaches the first threshold, it indicates that the accuracy of the layered identification result is high. The layered identification result can be directly identified as the biometric identification result of the target biometric feature. The layered fusion model can then stop the identification process and proceed to step four.
[0061] If the identification score of the latest layered identification result does not reach the first threshold, it means that the biometric identification result of the target biometric feature cannot be determined based on the latest layered identification result. Therefore, step three needs to be executed.
[0062] For example, the hierarchical fusion model consists of three layers, and the preset recognition order for each layer is: Layer 1, Layer 2, and Layer 3. In principle, the sum of the recognition scores in the output results of each layer should theoretically be 100%. The latest obtained hierarchical recognition result is layer recognition result 1 output by layer 1. It is determined that the recognition score "80%" in the latest obtained hierarchical recognition result reaches the first threshold "60%", indicating that the accuracy of layer recognition result 1 is high, and layer recognition result 1 can be directly identified as the biometric recognition result of the target biometric feature.
[0063] Step 3: Determine whether there are at least two target hierarchical identification results among the currently obtained results, wherein the identified objects in all target hierarchical identification results are the same, and the identification scores in all target hierarchical identification results reach the second threshold, which is less than the first threshold. If so, proceed to step 4; otherwise, proceed to step 5.
[0064] If the identification score in the latest layered identification result does not reach the first threshold, it means that the biometric identification result of the target biometric cannot be directly determined based on the latest layered identification result. Therefore, it is necessary to determine whether there are at least two target layered identification results among the currently obtained layered identification results.
[0065] The identified objects in all target hierarchical recognition results described here are the same, and the recognition scores in all target hierarchical recognition results reach the second threshold, which is less than the first threshold. The principle for setting the second threshold is to ensure that when the recognition scores in all target hierarchical recognition results are taken at the second threshold, the sum of the recognition scores in the target hierarchical recognition results is greater than half of the sum of the recognition scores in the hierarchical recognition results output by each layer. For example, the number of layers in the hierarchical fusion model is 4, and the second threshold is 30%. That is to say, when there are two hierarchical recognition results with recognition scores reaching 30% in the currently obtained hierarchical recognition results, it is determined that the condition for the hierarchical fusion model to stop recognition processing is met.
[0066] If it is determined that there are at least two target layer recognition results in the currently obtained layer recognition results, it means that the biometric recognition result of the target biometric can be determined based on the currently obtained layer recognition results. The layer fusion model can stop the recognition process and proceed to step four.
[0067] If it is determined that there are no at least two target layer recognition results in the currently obtained layer recognition results, it means that the biometric recognition result of the target biometrics cannot be determined based on the currently obtained layer recognition results. Therefore, step five needs to be executed.
[0068] Step 4: Determine the conditions that meet the criteria for the hierarchical fusion model to stop recognition processing, stop calling the hierarchical fusion model, and end the current process.
[0069] Once the conditions for stopping the hierarchical fusion model to perform identification processing are met, the hierarchical fusion model can be stopped from being invoked, thereby improving the speed of biometric identification by the hierarchical fusion model.
[0070] Step 5: Determine whether the latest obtained layer recognition result is the layer recognition result of the last layer output in the preset recognition order in the layer fusion model; if yes, end the current process; otherwise, execute step 1.
[0071] If it is determined that the latest layer recognition result is the layer recognition result of the last layer in the preset recognition order in the layer fusion model, it means that all layers in the layer fusion model have completed biometric recognition. The biometric recognition result needs to be determined based on the layer recognition results of all layers.
[0072] If it is determined that the latest layer recognition result is not the layer recognition result of the last layer in the preset recognition order in the layer fusion model, the layer fusion model is called again to perform biometric recognition so that the layer immediately following the current output layer recognition result can continue to recognize the corresponding image block group.
[0073] After completing steps one through five above, and when it is determined that the conditions for the hierarchical fusion model to stop recognition processing are met, it is necessary to determine the biometric recognition result based on the currently obtained hierarchical recognition results. The specific methods for determining the biometric recognition result are described below, and these methods include the following:
[0074] The first approach is to determine the latest layered identification result as a biometric identification result when the identification score reaches the first threshold.
[0075] When it is determined that the recognition score of the latest layer recognition result reaches the first threshold, it means that the latest layer recognition result has the highest recognition accuracy. Even if the layer that is later in the preset recognition order is used for biometric recognition, the accuracy of the resulting layer recognition result will not be higher than the accuracy of the latest layer recognition result. Therefore, the latest layer recognition result is determined as the biometric recognition result.
[0076] The second approach is to determine that the latest layered identification result does not reach the first threshold, but at least two target layered identification results exist in the current layered identification results. In this case, the target layered identification result with the highest identification score among the two target layered identification results is identified as the biometric identification result.
[0077] In principle, all target hierarchical identification results contain the same target, and each target hierarchical identification result can be used as a biometric identification result. However, in order to ensure the accuracy of the output biometric identification results, the target hierarchical identification result with the highest identification score among all target hierarchical identification results is determined as the biometric identification result of the target.
[0078] The third approach involves identifying the layered recognition result from all layers in the layered fusion model, and then determining the layered recognition result with the highest recognition score among the currently obtained layered recognition results as the biometric recognition result.
[0079] The recognition score reflects the probability that the image corresponding to the hierarchical recognition result is an image with the biometric features of the target object. Therefore, the higher the recognition score, the higher the probability that the image corresponding to the hierarchical recognition result is an image with the biometric features of the target object. Thus, the hierarchical recognition result with the highest recognition score among the currently obtained hierarchical recognition results is determined as the biometric recognition result.
[0080] The fourth method involves all layers in the hierarchical fusion model having output hierarchical recognition results, and the differences between the recognition scores in the currently obtained hierarchical recognition results being within a preset difference range. Hierarchical recognition results with the same recognition object are grouped together, and the hierarchical recognition result with the highest recognition score in the group with the most hierarchical recognition results is determined as the biometric recognition result.
[0081] For example, in the hierarchical fusion model, the hierarchical identification result 1 of hierarchical 1 is "Zhang San, 24%", the hierarchical identification result 2 of hierarchical 2 is "Zhang San, 25%", the hierarchical identification result 3 of hierarchical 3 is "Li Si, 25%", and the hierarchical identification result 4 of hierarchical 4 is "Wang Wu, 26%". It can be seen that the differences between the identification scores in the currently obtained hierarchical identification results are all within the preset difference range (-3%, 3%). The hierarchical identification results with the same identification object are grouped into one group: "Hierarchical identification result 1 and hierarchical identification result 2 are grouped into group 1, hierarchical identification result 3 is grouped into group 2, and hierarchical identification result 4 is grouped into group 3". The hierarchical identification result "hierarchical identification result 2" with the highest identification score in the group "group 1" with the most hierarchical identification results is determined as the biometric identification result.
[0082] The biometric recognition method provided in this invention, when biometric recognition is required, divides multiple images of different modalities of the target biometric feature into image block groups. Then, a hierarchical fusion model composed of multiple layers is invoked to perform recognition processing on each image block group. One layer in the hierarchical fusion model is used to recognize one image block group, and each layer has a preset recognition order. When the hierarchical fusion model performs biometric recognition, after any layer recognizes its corresponding image block group and outputs its hierarchical recognition result, the layer immediately following it begins to recognize its corresponding image block group. When the condition for the hierarchical fusion model to stop recognition processing is determined, the biometric recognition result is determined based on the currently obtained hierarchical recognition results. Therefore, this invention, in order to reduce the impact of environmental factors such as lighting and biometric acquisition angle on the accuracy of biometric recognition, performs biometric recognition on image block groups after dividing multiple images of different modalities of the target biometric feature into blocks. Furthermore, to improve the speed of biometric recognition, once all the currently obtained hierarchical recognition results meet the requirements for generating the biometric recognition result of the target biometric feature, the hierarchical fusion model can stop biometric recognition. This eliminates the need for the hierarchical fusion model to complete the recognition of all image blocks; the biometric recognition result can be determined based on the currently obtained hierarchical recognition results, thereby reducing the biometric recognition time and accelerating the biometric recognition speed. In summary, the solution provided by the embodiments of the present invention can improve both the accuracy and speed of biometric recognition.
[0083] Furthermore, according to Figure 1 The method shown in the figure, another embodiment of the present invention also provides a biometric identification method, such as Figure 2 As shown, the method mainly includes:
[0084] 201. Acquire at least two images of the target biometric features, wherein each image has a different modality.
[0085] In practical applications, at least two images of the target biomarker are acquired using multispectral equipment, with each image corresponding to a different spectrum.
[0086] For example, palm prints are chosen as the target biometric feature because they are rich in features, easily obtainable, and unique. Figure 3 The multispectral palmprint acquisition device shown collects palmprints. Figure 3In the diagram, A1 is the spectral control element, A2 is the multispectral lamp, A3 is the image acquisition device, and B represents the placement of the human hand. When acquiring palmprint images, the palm must be positioned opposite the image acquisition device. After each palmprint image is acquired, the spectral control element controls the multispectral lamp to change its spectrum. After the multispectral device acquires the palmprint, three images are obtained: Image 1, Image 2, and Image 3. Image 1 corresponds to the blue spectral mode, meaning it was acquired under the blue spectrum. Image 2 corresponds to the green spectral mode, meaning it was acquired under the green spectrum. Image 3 corresponds to the red spectral mode, meaning it was acquired under the red spectrum.
[0087] 202. Divide at least two images of the target biometric features into blocks to form image block groups for each image.
[0088] 203. Call the layered fusion model composed of multiple layers to perform recognition processing on each of the image block groups. Here, one layer is used to identify one image block group, and each layer has a preset recognition order. After any layer identifies the corresponding image block group and outputs the layer recognition result, the layer immediately following it in the sequence begins to identify the corresponding image block group.
[0089] 204. When it is determined that the conditions for stopping the recognition process of the hierarchical fusion model are met, the biometric recognition result is determined based on the currently obtained hierarchical recognition result.
[0090] The specific explanations of steps 202 to 204 above are as follows: Figure 1 The steps 101 and 103 are basically the same, so they will not be repeated here.
[0091] Furthermore, based on the above method embodiments, another embodiment of the present invention also provides a biometric identification device, such as... Figure 4 As shown, the device includes:
[0092] Segmentation unit 31 is used to segment at least two images of the target biometric feature to form image segmentation groups for each image, wherein each image has a different modality;
[0093] Calling unit 32 is used to call a layered fusion model composed of multiple layers to perform recognition processing on each of the image block groups. Here, one layer is used to identify one image block group, and each layer has a preset recognition order. After any layer identifies the corresponding image block group and outputs the layer recognition result, the layer immediately following it in the sequence begins to identify the corresponding image block group.
[0094] The determining unit 33 is used to determine the biometric recognition result based on the currently obtained hierarchical recognition result when it is determined that the condition for stopping the recognition process of the hierarchical fusion model is met.
[0095] The biometric recognition device provided in this invention, when biometric recognition is required, divides multiple images of different modalities of the target biometric feature into image block groups. Then, a hierarchical fusion model composed of multiple layers is invoked to perform recognition processing on each image block group. One layer in the hierarchical fusion model is used to recognize one image block group, and each layer has a preset recognition order. When the hierarchical fusion model performs biometric recognition, after any layer recognizes its corresponding image block group and outputs its hierarchical recognition result, the layer immediately following it begins to recognize its corresponding image block group. When the condition for the hierarchical fusion model to stop recognition processing is met, the biometric recognition result is determined based on the currently obtained hierarchical recognition results. Therefore, this invention, in order to reduce the impact of environmental factors such as lighting and biometric acquisition angle on the accuracy of biometric recognition, performs biometric recognition on image block groups after dividing multiple images of different modalities of the target biometric feature into blocks. Furthermore, to improve the speed of biometric recognition, once all the currently obtained hierarchical recognition results meet the requirements for generating the biometric recognition result of the target biometric feature, the hierarchical fusion model can stop biometric recognition. This eliminates the need for the hierarchical fusion model to complete the recognition of all image blocks; the biometric recognition result can be determined based on the currently obtained hierarchical recognition results, thereby reducing the biometric recognition time and accelerating the biometric recognition speed. In summary, the solution provided by the embodiments of the present invention can improve both the accuracy and speed of biometric recognition.
[0096] Optional, such as Figure 5 As shown, the determining unit 33 includes:
[0097] The first determining module 331 is used to determine, when it is determined that the recognition score of the latest obtained hierarchical recognition result reaches a first threshold, that the condition for stopping the recognition processing of the hierarchical fusion model is met. The hierarchical recognition result includes a recognition score and a target object. The recognition score is used to reflect the probability that the image corresponding to the hierarchical recognition result is an image of the biometric features of the target object.
[0098] Optional, such as Figure 5 As shown, the determining unit 33 includes:
[0099] The second determining module 332 is used to determine the latest obtained hierarchical identification result as the biometric identification result.
[0100] Optional, such as Figure 5 As shown, the determining unit 33 includes:
[0101] The third determining module 333 is used to determine whether there are at least two target layered identification results in the currently obtained layered identification results when the identification score of the latest obtained layered identification result does not reach the first threshold; and to determine that the condition for the layered fusion model to stop identification processing is met when at least two target layered identification results are found in the currently obtained layered identification results. Wherein, the identification objects in all target layered identification results are the same, and the identification scores in all target layered identification results reach the second threshold, which is less than the first threshold.
[0102] Optional, such as Figure 5 As shown, the determining unit 33 includes:
[0103] The fourth determining module 334 is used to determine the target layer recognition result with the highest recognition score among the at least two target layer recognition results as the biometric recognition result.
[0104] Optional, such as Figure 5 As shown, the determining unit 33 includes:
[0105] The fifth determining module 335 is used to determine the layer recognition result with the highest recognition score among the currently obtained layer recognition results after all layers in the layer fusion model have output layer recognition results as the biometric recognition result. The layer recognition result includes a recognition score and a target object, whereby the recognition score reflects the probability that the image corresponding to the layer recognition result is an image representing the biometric features of the target object.
[0106] Optional, such as Figure 5 As shown, the determining unit 33 includes:
[0107] The sixth determining module 336 is used to group the layered recognition results with the same recognition object into a group when all layers in the layered fusion model have output layered recognition results and the differences between the recognition scores in the currently obtained layered recognition results are all within a preset difference range, and to determine the layered recognition result with the highest recognition score in the group with the most layered recognition results as the biometric recognition result; wherein, the layered recognition result includes a recognition score and a target object, and the recognition score is used to reflect the probability that the image corresponding to the layered recognition result is an image of the biometric features of the target object.
[0108] Optional, such as Figure 5As shown, the calling unit 32 is specifically used to determine the image block group corresponding to each layer. In the preset recognition order, the layer that is ranked earlier has fewer image blocks in its corresponding image block group. The calling unit 32 sequentially identifies the image block group corresponding to each layer.
[0109] Optional, such as Figure 5 As shown, the device further includes:
[0110] The acquisition unit 34 is used to acquire at least two images of the target biofeature using a multispectral device, wherein each image corresponds to a different spectrum.
[0111] For a detailed explanation of the methods used in the operation of each functional module of the biometric identification device provided in this embodiment of the invention, please refer to [link to relevant documentation]. Figures 1-2 The detailed explanations of the corresponding methods in the method embodiments will not be repeated here.
[0112] Furthermore, according to the above embodiments, another embodiment of the present invention also provides a computer-readable storage medium, the storage medium including a stored program, wherein the program controls the device where the storage medium is located to execute [the program]. Figures 1-2 The aforementioned biometric identification method.
[0113] Furthermore, according to the above embodiments, another embodiment of the present invention also provides an electronic device, the electronic device comprising:
[0114] Memory, used to store programs;
[0115] A processor, coupled to the memory, is used to run the program to execute... Figures 1-2 The aforementioned biometric identification method.
[0116] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.
[0117] It is understood that the relevant features in the above methods and apparatus can be referenced interchangeably. Furthermore, the terms "first," "second," etc., in the above embodiments are used to distinguish between embodiments and do not represent the superiority or inferiority of any particular embodiment.
[0118] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.
[0119] The algorithms and displays provided herein are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used in conjunction with the teachings herein. The required structure for constructing such systems is apparent from the above description. Furthermore, this invention is not directed to any particular programming language. It should be understood that the contents of the invention described herein can be implemented using various programming languages, and the above description of specific languages is for the purpose of disclosing the best mode of implementation of the invention.
[0120] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.
[0121] Furthermore, those skilled in the art will understand that although some embodiments described herein include certain features but not others included in other embodiments, combinations of features from different embodiments are intended to be within the scope of the invention and form different embodiments. For example, in the claims, any of the claimed embodiments can be used in any combination.
[0122] The various component embodiments of the present invention can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that microprocessors or digital signal processors (DSPs) can be used in practice to implement some or all of the functions of some or all of the components in the method, apparatus, and framework for operating the deep neural network model according to embodiments of the present invention. The present invention can also be implemented as a device or apparatus program (e.g., a computer program and computer program product) for performing part or all of the methods described herein. Such programs implementing the present invention can be stored on a computer-readable medium, or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.
[0123] It should be noted that the above embodiments are illustrative of the invention and not restrictive, and that those skilled in the art can devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses should not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The invention can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names.
Claims
1. A biometric identification method, characterized in that, The method includes: At least two images of the target biometric feature are segmented to form image block groups for each image, wherein each image has a different modality and corresponds to a different spectrum; A layered fusion model composed of multiple layers is invoked to identify each of the image block groups. Each layer is used to identify one image block group, and each layer has a preset identification order. After any layer identifies its corresponding image block group and outputs the layered identification result, the layer immediately following it begins to identify its corresponding image block group. Since there is a set modality identification order between modalities, the invocation of the layered fusion model to identify each of the image block groups includes: determining the corresponding image block group for each layer based on the modality identification order; sequentially invoking each layer to identify its corresponding image block group; and invoking the layered fusion model to identify each of the image block groups includes: determining the image block group corresponding to each layer, wherein the layer ranked earlier in the preset identification order has fewer image blocks in its corresponding image block group; and sequentially invoking each layer to identify its corresponding image block group. When it is determined that the conditions for stopping the recognition process of the hierarchical fusion model are met, the biometric recognition result is determined based on the currently obtained hierarchical recognition result; The conditions for determining whether the hierarchical fusion model can stop recognition processing include: when it is determined that the recognition score of the latest obtained hierarchical recognition result reaches a first threshold, the conditions for the hierarchical fusion model to stop recognition processing are met, wherein the hierarchical recognition result includes a recognition score and a target object, and the recognition score is used to reflect the probability that the image corresponding to the hierarchical recognition result is an image of the biometric features of the target object; when it is determined that the recognition score of the latest obtained hierarchical recognition result does not reach the first threshold, it is determined whether there are at least two target hierarchical recognition results in the currently obtained hierarchical recognition results; wherein the recognition objects in all target hierarchical recognition results are the same, and the recognition scores in all target hierarchical recognition results reach a second threshold, the second threshold being less than the first threshold; when it is determined that there are at least two target hierarchical recognition results in the currently obtained hierarchical recognition results, the conditions for the hierarchical fusion model to stop recognition processing are met, and the hierarchical fusion model is stopped.
2. The method according to claim 1, characterized in that, Based on the currently obtained hierarchical identification results, the biometric identification results are determined, including: The latest obtained hierarchical identification result is determined as the biometric identification result.
3. The method according to claim 1, characterized in that, Based on the currently obtained hierarchical identification results, the biometric identification results are determined, including: The target layer recognition result with the highest recognition score among the at least two target layer recognition results is determined as the biometric recognition result.
4. The method according to claim 1, characterized in that, Based on the currently obtained hierarchical identification results, the biometric identification results are determined, including: In the hierarchical fusion model, all layers have output hierarchical identification results. The hierarchical identification result with the highest identification score among the currently obtained hierarchical identification results is determined as the biometric identification result. or, In the hierarchical fusion model, all layers have output hierarchical recognition results, and the differences between the recognition scores in the currently obtained hierarchical recognition results are all within the preset difference range. The hierarchical recognition results with the same recognition object are grouped together, and the hierarchical recognition result with the highest recognition score in the group with the most hierarchical recognition results is determined as the biometric recognition result. The hierarchical recognition result includes a recognition score and a target object. The recognition score is used to reflect the probability that the image corresponding to the hierarchical recognition result is an image of the biometric features of the target object.
5. The method according to claim 1, characterized in that, The method further includes: At least two images of the target biofeature are acquired using a multispectral device, wherein each image corresponds to a different spectrum.
6. A biometric identification device, characterized in that, The device includes: A segmentation unit is used to segment at least two images of a target biometric feature into blocks, forming image block groups for each image, wherein each image has a different modality and corresponds to a different spectrum; The calling unit is used to call a layered fusion model composed of multiple layers to perform recognition processing on each of the image block groups. Each layer is used to identify one image block group, and each layer has a preset recognition order. After any layer identifies its corresponding image block group and outputs the layered recognition result, the layer immediately following it in the sequence begins to identify its corresponding image block group. If there is a set modality recognition order between modalities, then calling the layered fusion model composed of multiple layers to perform recognition processing on each of the image block groups includes: determining the corresponding image block group for each layer based on the modality recognition order; sequentially calling each layer to identify its corresponding image block group; and calling the layered fusion model composed of multiple layers to perform recognition processing on each of the image block groups includes: determining the image block group corresponding to each layer, wherein the layer ranked earlier in the preset recognition order has fewer image blocks in its corresponding image block group; and sequentially calling each layer to identify its corresponding image block group. The determining unit is used to determine the biometric identification result based on the currently obtained hierarchical identification result when it is determined that the condition for stopping the identification process of the hierarchical fusion model is met. The determining unit is configured to, when determining that the recognition score of the latest obtained layered recognition result reaches a first threshold, determine that the condition for the layered fusion model to stop recognition processing is met, wherein the layered recognition result includes a recognition score and a target object, and the recognition score is used to reflect the probability that the image corresponding to the layered recognition result is an image of the biometric features of the target object; when determining that the recognition score of the latest obtained layered recognition result does not reach the first threshold, determine whether there are at least two target layered recognition results in the currently obtained layered recognition results; wherein the recognition objects in all target layered recognition results are the same, and the recognition scores in all target layered recognition results reach a second threshold, the second threshold being less than the first threshold; when determining that there are at least two target layered recognition results in the currently obtained layered recognition results, determine that the condition for the layered fusion model to stop recognition processing is met, and stop calling the layered fusion model.
7. A computer-readable storage medium, characterized in that, The storage medium includes a stored program, wherein, when the program is executed, it controls the device containing the storage medium to perform the biometric identification method according to any one of claims 1 to 5.
8. An electronic device, characterized in that, The electronic device includes: Memory, used to store programs; A processor, coupled to the memory, is used to run the program to perform the biometric identification method according to any one of claims 1 to 5.
Citation Information
Patent Citations
Target recognition method
CN112149687A