WO2023083051A1 - 生物特征识别方法、装置、设备和存储介质 - Google Patents
生物特征识别方法、装置、设备和存储介质 Download PDFInfo
- Publication number
- WO2023083051A1 WO2023083051A1 PCT/CN2022/128883 CN2022128883W WO2023083051A1 WO 2023083051 A1 WO2023083051 A1 WO 2023083051A1 CN 2022128883 W CN2022128883 W CN 2022128883W WO 2023083051 A1 WO2023083051 A1 WO 2023083051A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- feature
- sorting
- matching algorithms
- matrix
- feature matching
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims abstract description 60
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 192
- 239000011159 matrix material Substances 0.000 claims abstract description 125
- 238000004364 calculation method Methods 0.000 claims abstract description 14
- 230000015654 memory Effects 0.000 claims description 16
- 238000004590 computer program Methods 0.000 claims description 7
- 238000012545 processing Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 2
- 238000011426 transformation method Methods 0.000 claims description 2
- 239000000523 sample Substances 0.000 description 41
- 238000005516 engineering process Methods 0.000 description 13
- 238000013528 artificial neural network Methods 0.000 description 7
- 238000010276 construction Methods 0.000 description 7
- 238000012360 testing method Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 238000011156 evaluation Methods 0.000 description 3
- 210000000887 face Anatomy 0.000 description 2
- 210000003811 finger Anatomy 0.000 description 2
- 210000000554 iris Anatomy 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000007619 statistical method Methods 0.000 description 2
- 210000003462 vein Anatomy 0.000 description 2
- 102100024607 DNA topoisomerase 1 Human genes 0.000 description 1
- 101000830681 Homo sapiens DNA topoisomerase 1 Proteins 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
Definitions
- the present application relates to the technical field of information processing, and in particular to a biometric identification method, device, equipment and storage medium.
- Biometrics technology is closely combined with high-tech means such as computer, optics, acoustics, biosensors and biostatistics principles, and uses the inherent physiological characteristics of the human body such as fingerprints, faces, irises, and finger veins for personal identification.
- Biometric identification belongs to the 1:N comparison method. People's life and work have brought convenience. Nevertheless, in the practical application of 1:N comparison, problems of misrecognition, false alarm and false negative often occur, which bring troubles and risks to the system application.
- the purpose of the embodiments of the present application is to provide a biometric feature identification method, device, device, and storage medium, which can improve the reliability of biometric feature identification through the comparison and ranking of multiple matching algorithms and the sorting weight of each algorithm.
- the first aspect of the embodiment of the present application provides a biometric identification method, including: obtaining a biometric sample to be identified; using a variety of feature matching algorithms to compare the biometric sample with the feature templates in the feature template library one by one Yes, obtain the corresponding set of candidate feature templates under each of the feature matching algorithms; perform weighted calculations on the ordering of each of the set of candidate feature templates according to the sorting weight of each of the feature matching algorithms in the weight matrix; according to the weighted After sorting each set of candidate feature templates, the final target object corresponding to the biometric sample is determined.
- multiple feature matching algorithms are used to compare the biological feature samples with the feature templates in the feature template library one by one to obtain a set of candidate feature templates corresponding to each of the feature matching algorithms , including: using the first feature matching algorithm among the multiple feature matching algorithms, comparing the biological feature sample with the feature templates in the feature template library one by one, and selecting a first set of candidate feature templates; Using each of the remaining feature matching algorithms in the plurality of feature matching algorithms respectively, comparing the biometric sample with the feature templates in the first candidate feature template set one by one to obtain the feature matching algorithm for each The corresponding second set of candidate feature templates.
- the biological feature recognition method before the weighted calculation of the ranking of each of the candidate feature template sets according to the sorting weight of each of the feature matching algorithms in the weight matrix, the biological feature recognition method further The method includes: constructing the weight matrix of the plurality of feature matching algorithms respectively according to the recognition error rate of each of the candidate feature template sets, and the weight matrix includes the sorting weight of each feature matching algorithm.
- the identification error rate includes: the error identification selection rate of each of the candidate feature template sets; the following formula is used to construct the weight matrix of the various feature matching algorithms:
- m is the total number of the multiple feature matching algorithms
- m is a positive integer
- W is the weight matrix of the multiple feature matching algorithms
- w is the i-th algorithm corresponding to the multiple feature matching algorithms
- the sorting weight, FISRi is the misidentification selectivity rate of the candidate feature template set corresponding to the i-th algorithm among the multiple feature matching algorithms.
- the recognition error rate includes: the false recognition selection rate and the false recognition rejection rate of each of the candidate feature template sets; the following formula is used to construct the weight matrix of the various feature matching algorithms:
- m is the total number of the multiple feature matching algorithms
- m is a positive integer
- W is the weight matrix of the multiple feature matching algorithms
- w is the i-th algorithm corresponding to the multiple feature matching algorithms
- the sorting weight FISRi is the error identification selectivity rate of the candidate feature template set corresponding to the i-th algorithm among the various feature matching algorithms
- FIRRi is the error recognition selectivity rate corresponding to the i-th algorithm among the various feature matching algorithms.
- the false identification rejection rate of the set of candidate feature templates is the false identification rejection rate of the set of candidate feature templates.
- the weighted calculation of the sorting of each of the candidate feature template sets according to the sorting weights of each of the feature matching algorithms in the weight matrix includes: according to each of the feature matching According to the set of candidate feature templates corresponding to the algorithm, an initial sorting matrix corresponding to the various feature matching algorithms is generated; after each element of the initial sorting matrix is reciprocated, a reciprocal sorting matrix is generated; and the reciprocal sorting matrix is standardized Processing to obtain a standardized ranking matrix; multiplying the weight matrix by the standardized ranking matrix to obtain the final ranking matrix of all candidate feature templates corresponding to the biological feature samples under the multiple feature matching algorithms.
- the determining the final target object corresponding to the biometric sample according to each set of candidate feature templates after weighted sorting includes: arranging the elements in the final sorting matrix in descending order, Determining the candidate feature templates corresponding to the elements with a preset top ranking in the final sorting matrix as the final target object corresponding to the biometric sample.
- the second aspect of the embodiment of the present application provides a biometric identification device, including: an acquisition module, used to acquire a biometric sample to be identified; a comparison module, used to use a variety of feature matching algorithms to compare the biometric sample Comparing the feature templates in the feature template library one by one to obtain a set of candidate feature templates corresponding to each of the feature matching algorithms; the weighting module is used to perform a weighting according to the sorting weight of each of the feature matching algorithms in the weight matrix. The ranking of each set of candidate feature templates is weighted; the determination module is configured to determine the final target object corresponding to the biometric sample according to each set of candidate feature templates after weighted sorting.
- the comparison module is configured to: use the first feature matching algorithm among the multiple feature matching algorithms to perform one-by-one comparison between the biometric sample and the feature templates in the feature template library Comparing, selecting the first set of candidate feature templates; using each of the remaining feature matching algorithms in the multiple feature matching algorithms, respectively, comparing the biometric sample with the feature templates in the first set of candidate feature templates Compare one by one to obtain the second set of candidate feature templates corresponding to each feature matching algorithm.
- the biometric identification device further includes: a construction module for sorting each of the candidate feature template sets according to the sorting weights of each of the feature matching algorithms in the weight matrix Before doing the weighted calculation, according to the recognition error rate of each of the candidate feature template sets, the weight matrix of the multiple feature matching algorithms is constructed, and the weight matrix includes the sorting weight of each of the feature matching algorithms .
- the identification error rate includes: the error identification selection rate of each of the candidate feature template sets; the construction module is used to construct the weight matrix of the various feature matching algorithms using the following formula:
- m is the total number of the multiple feature matching algorithms
- m is a positive integer
- W is the weight matrix of the multiple feature matching algorithms
- w is the i-th algorithm corresponding to the multiple feature matching algorithms
- the sorting weight, FISRi is the misidentification selectivity rate of the candidate feature template set corresponding to the i-th algorithm among the multiple feature matching algorithms.
- the recognition error rate includes: the false recognition selection rate and the false recognition rejection rate of each of the candidate feature template sets; the construction module is used to construct the various feature matching algorithms using the following formula The weight matrix of :
- m is the total number of the multiple feature matching algorithms
- m is a positive integer
- W is the weight matrix of the multiple feature matching algorithms
- w is the i-th algorithm corresponding to the multiple feature matching algorithms
- the sorting weight FISRi is the error identification selectivity rate of the candidate feature template set corresponding to the i-th algorithm among the various feature matching algorithms
- FIRRi is the error recognition selectivity rate corresponding to the i-th algorithm among the various feature matching algorithms.
- the false identification rejection rate of the set of candidate feature templates is the false identification rejection rate of the set of candidate feature templates.
- the weighting module is configured to: generate an initial sorting matrix corresponding to the various feature matching algorithms according to the set of candidate feature templates corresponding to each of the feature matching algorithms; After taking the reciprocal of each element of , a reciprocal sorting matrix is generated; the reciprocal sorting matrix is standardized to obtain a standardized sorting matrix; the weight matrix is multiplied by the normalized sorting matrix to obtain the biological characteristic sample in the The final ranking matrix corresponding to all candidate feature templates under the above-mentioned multiple feature matching algorithms.
- the determining module is configured to: arrange the elements in the final sorting matrix in descending order, and determine the candidate feature templates corresponding to the elements in the final ranking matrix with the top preset rankings as the The final target object corresponding to the biometric sample.
- the third aspect of the embodiments of the present application provides an electronic device, including: a memory for storing computer programs; a processor for executing the computer programs, so as to realize any one of the first aspects of the embodiments of the present application Methods.
- the fourth aspect of the embodiment of the present application provides a non-transitory electronic device-readable storage medium, including: a program, which, when run by the electronic device, enables the electronic device to execute any of the above aspects of the first aspect of the embodiment of the present application.
- a method of an embodiment is a non-transitory electronic device-readable storage medium, including: a program, which, when run by the electronic device, enables the electronic device to execute any of the above aspects of the first aspect of the embodiment of the present application.
- the biometric identification method, device, equipment, and storage medium provided by this application compare the biometric samples to be identified with the feature templates in the feature template library one by one through a variety of feature matching algorithms, so as to obtain the following characteristics of each feature matching algorithm.
- the corresponding candidate feature template set, and according to the recognition error rate under each candidate feature template set construct the weight matrix of multiple feature matching algorithms.
- each candidate The sorting of the feature template set is weighted. In this way, the comparison and ranking of multiple matching algorithms are used, and the sorting weight of each algorithm is considered, thereby improving the reliability of biometric identification.
- FIG. 1 is a schematic structural diagram of an electronic device according to an embodiment of the present application.
- Fig. 2 is a schematic flowchart of a biometric identification method according to an embodiment of the present application.
- Fig. 3 is a schematic flowchart of a biometric identification method according to an embodiment of the present application.
- Fig. 4 is a schematic structural diagram of a biometric identification device according to an embodiment of the present application.
- Fig. 5 is a schematic diagram of an internal structure of a device for biometric identification according to an embodiment of the present application.
- this embodiment provides an electronic device 1 , including: at least one processor 11 and a memory 12 .
- a processor is taken as an example.
- Processor 11 and memory 12 are connected via bus 10 .
- the memory 12 stores computer instructions that can be executed by the processor 11, and the computer instructions are executed by the processor 11, so that the electronic device 1 can execute all or part of the procedures of the methods in the following embodiments, so that through various matching algorithms Compare rankings and ranking weights for each algorithm to improve biometric reliability.
- the electronic device 1 may be a mobile phone, a tablet computer, a notebook computer, a desktop computer, or a server composed of multiple computers.
- FIG. 2 is a biometric identification method according to an embodiment of the present application. This method can be executed by the electronic device 1 shown in FIG. Reliability of feature recognition. The method comprises the steps of:
- Step 210 Obtain a biometric sample to be identified.
- the biometric samples to be identified refer to the inherent physiological characteristics of the human body, such as fingerprints, faces, irises, and finger veins.
- the fingerprint image acquisition method can be optical fingerprint acquisition technology, capacitive sensor fingerprint acquisition technology, temperature sensor fingerprint acquisition technology, ultrasonic fingerprint acquisition technology, electromagnetic wave fingerprint acquisition technology wait.
- fingerprint feature samples to be identified can be obtained by using image acquisition software, such as Anguli, ABIS multi-modal biological identification unified platform, and the like.
- Step 220 Using multiple feature matching algorithms, comparing the biometric sample with the feature templates in the feature template library one by one to obtain a set of candidate feature templates corresponding to each feature matching algorithm.
- the feature matching algorithm refers to an algorithm for matching the features of the biometric sample with the feature templates in the feature template library.
- the face recognition algorithm includes a face recognition method based on geometric features, a face recognition method based on subspace, a face recognition method based on statistics, and an elastic graph matching method.
- face recognition method based on hybrid model face recognition method based on 3D model and face recognition method based on deep neural network, etc.
- the face recognition method based on statistics includes the face recognition method based on KL transform, the face recognition method based on hidden Markov model, and the like.
- multiple feature matching algorithms can be different types of algorithms, for example, the first is a face recognition method based on a deep neural network, the second is a subspace-based face recognition method, and the third is a face recognition method based on a deep neural network.
- the first is a face recognition method based on a deep neural network
- the second is a subspace-based face recognition method
- the third is a face recognition method based on a deep neural network.
- the face recognition method based on deep neural network is used as the first feature matching algorithm, and the face to be recognized is compared with the face feature template one by one using the face recognition method based on deep neural network to obtain the matching score, and The obtained matching scores are sorted in descending order; at the same time, the feature templates corresponding to the top k bits are used as the first set of candidate feature templates.
- k is a custom length.
- the remaining face recognition algorithms are respectively used to compare the face samples to be recognized with the feature templates in the first candidate feature template set one by one to obtain the second candidate feature template set corresponding to each feature matching algorithm.
- multiple feature matching algorithms may be of the same type, such as algorithms based on neural networks, but feature matching algorithms may be different due to different neural networks.
- Step 230 Construct weight matrices of multiple feature matching algorithms according to the recognition error rate of each candidate feature template set.
- the weight matrix includes the sorting weights of each feature matching algorithm, and the sorting weights of each feature matching algorithm can be statistically determined in advance to form a weight matrix.
- the feature template set refers to a registration set, that is, a set of registration templates used for biometric identification (1:N comparison).
- the performance of the biometric recognition algorithm is reflected in indicators such as false recognition selection rate (FISR), false recognition rejection rate (FIRR), preferred recognition rate (TOP1), and top k selected recognition rate (TOP k).
- FISR false recognition selection rate
- FIRR false recognition rejection rate
- TOP1 preferred recognition rate
- TOP k top k selected recognition rate
- the evaluation of algorithm performance indicators is carried out on a certain test sample set.
- the test sample set can be divided into a registration set and a probe set; wherein, the probe set is a collection of samples (also called probes) to be identified for algorithm performance evaluation.
- the sample to be identified may or may not have a matching template in the registration set.
- the number of samples in the registration set and the probe set may or may not be the same.
- the number of samples in the test set is generally constructed according to the actual application data scale, such as: 10,000, 100,000, 1 million, 10 million, etc.
- the quality of algorithm performance indicators is related to the quality of test samples.
- the quality of test samples in the sample set should follow the sample quality standards corresponding to the biometric modality.
- the recognition error rate includes a false recognition selection rate (FISR) and a false recognition rejection rate (FIRR).
- FISR index is related to the number of samples in the registration set N, the length of the candidate list K, and the comparison threshold T, etc., and can be expressed as FISR(N,K,T).
- the statistical method of the FISR index is that in the search that cannot match the target person, there are candidates whose scores exceed the threshold T and rank within K in the comparison result list (all are wrong matches), FISR(N,K,T ) is the ratio of the number of candidates to the number of unmatched samples in the probe set.
- N is the number of samples in the registration set
- K is the length of the candidate list
- T is the comparison threshold
- C 2 is the sum of the number of candidates whose scores exceed the threshold T and are ranked within K in the first round of comparison search
- C 1 is the number of unmatched samples in the set of samples to be identified.
- the FIRR index is related to the number of samples in the registration set N, the length of the candidate list K, and the comparison threshold T, which can be expressed as FIRR(N,K,T).
- the statistical method is that in the search that can match the target person, in the comparison result list, there are wrongly rejected persons whose comparison score between the probe and the corresponding template does not exceed the threshold T or whose ranking is outside the K position (it should be matched but not matched) losers), FIRR(N,K,T) is the ratio of the number of false rejections to the number of matching samples in the probe set.
- C 4 is the number of wrongly rejected samples whose scores do not exceed the threshold T or rank outside the K position in C 3 rounds of comparison search;
- C 3 is the number of matching samples in the set of samples to be identified.
- m is the total number of multiple feature matching algorithms, and m is a positive integer; W is the weight matrix of multiple feature matching algorithms; w i is the sorting weight corresponding to the i-th algorithm in multiple feature matching algorithms; FISRi is FIRRi is the false recognition rejection rate of the candidate feature template set corresponding to the i-th algorithm among multiple feature matching algorithms.
- step 230 needs to occur before step 240, and the weight matrix can also be determined after the various feature matching algorithms used are determined. In actual biometric identification, as long as the feature matching algorithm remains unchanged, there is no need to construct the weight matrix each time, so step 230 can also occur before step 210-step 220, or at the same time. This embodiment does not limit the order of occurrence between step 230 and step 210 and step 220 .
- Step 240 Perform weighted calculation on the ranking of each candidate feature template set according to the ranking weight of each feature matching algorithm in the weight matrix.
- the ranking of each candidate feature template set refers to a matrix composed of ranking scores of each candidate feature template.
- Step 250 Determine the final target object corresponding to the biometric sample according to each set of candidate feature templates after weighted sorting.
- the final target object may be the target object ranked first after weighted sorting, or it may be a result set of candidates returned in descending order according to the set length.
- Fig. 3 is a biometric identification method according to an embodiment of the present application, which includes the following steps:
- Step 310 Obtain a biometric sample to be identified. For details, refer to the description of step 210 in the foregoing embodiments.
- Step 320 Using a variety of feature matching algorithms, comparing the biometric sample with the feature templates in the feature template library one by one to obtain a set of candidate feature templates corresponding to each feature matching algorithm. For details, refer to the description of step 220 in the foregoing embodiments.
- Step 330 Construct weight matrices of various feature matching algorithms according to the recognition error rate of each candidate feature template set. For details, refer to the description of step 230 in the foregoing embodiments.
- Step 340 According to the set of candidate feature templates corresponding to each feature matching algorithm, generate initial ranking matrices corresponding to multiple feature matching algorithms.
- Step 350 After reciprocating each element of the initial sorting matrix, generate a reciprocal sorting matrix.
- the initial sorting matrix refers to generating a sorting matrix (r ij ) m ⁇ k according to each round of comparison of candidate lists, and the specific form is as follows:
- m is that the biometric sample to be identified has undergone m rounds of comparison
- k is the length of the candidate set
- r ij is the matching score of the biometric sample to be identified that ranks j after the i round of matching.
- Step 360 Standardize the reciprocal sort matrix to obtain a normalized sort matrix.
- the standardization process refers to performing z-score transformation on the elements in the reciprocal sorting matrix, so as to obtain the standardized sorting matrix (z ij ) m ⁇ k , the specific form is as follows:
- the z-score transformation method is as follows:
- b ij is the value of row i and column j in the reciprocal sorting matrix
- ⁇ is the average value of b ij
- ⁇ is the standard deviation of b ij .
- Step 370 Multiply the weight matrix and the standardized ranking matrix to obtain the final ranking matrix of all candidate feature templates corresponding to the biological feature samples under multiple feature matching algorithms.
- Step 380 Determine the final target object corresponding to the biometric sample according to each weighted and sorted set of candidate feature templates. For details, refer to the description of step 250 in the foregoing embodiments.
- FIG. 4 is a biometric identification device 400 provided by an embodiment of the present application. This device can be applied to the electronic device 1 shown in FIG. Sorting weights improve the reliability of biometric identification.
- the device includes: an acquisition module 401, a comparison module 402, a weighting module 403 and a determination module 404, and the principle relationship of each module is as follows:
- the acquisition module 401 is configured to acquire a biometric sample to be identified.
- the comparison module 402 is used to compare the biometric sample with the feature templates in the feature template library one by one by using multiple feature matching algorithms, so as to obtain a set of candidate feature templates corresponding to each feature matching algorithm;
- the weighting module 403 is used to perform weighted calculation on the sorting of each candidate feature template set according to the sorting weight of each feature matching algorithm in the weight matrix;
- the determining module 404 is configured to determine the final target object corresponding to the biometric sample according to each weighted and sorted set of candidate feature templates.
- the comparison module 402 is also used to: use the first feature matching algorithm among multiple feature matching algorithms to compare the biological feature samples with the feature templates in the feature template library one by one, and select The first set of candidate feature templates; each of the remaining feature matching algorithms in the multiple feature matching algorithms is used to compare the biometric samples with the feature templates in the first set of candidate feature templates one by one, and the results of each feature matching algorithm are obtained. The corresponding second set of candidate feature templates.
- the biometric feature recognition device 400 further includes: a construction module 405, used to perform weighted calculations on the sorting of each candidate feature template set according to the sorting weight of each feature matching algorithm in the weight matrix , according to the recognition error rate of each candidate feature template set, construct the weight matrix of multiple feature matching algorithms, where the weight matrix includes the sorting weight of each feature matching algorithm.
- a construction module 405 used to perform weighted calculations on the sorting of each candidate feature template set according to the sorting weight of each feature matching algorithm in the weight matrix , according to the recognition error rate of each candidate feature template set, construct the weight matrix of multiple feature matching algorithms, where the weight matrix includes the sorting weight of each feature matching algorithm.
- the recognition error rate includes: the wrong recognition selection rate of each candidate feature template set; the construction module 405 is also used to construct the weight matrix of various feature matching algorithms using the following formula:
- m is the total number of multiple feature matching algorithms, and m is a positive integer; W is the weight matrix of multiple feature matching algorithms; w i is the sorting weight corresponding to the i-th algorithm in multiple feature matching algorithms; The misidentification selection rate of the candidate feature template set corresponding to the i-th algorithm in one feature matching algorithm.
- the recognition error rate includes: the false recognition selection rate and false recognition rejection rate of each candidate feature template set; the construction module 405 is also used to construct a weight matrix of various feature matching algorithms using the following formula:
- m is the total number of multiple feature matching algorithms, and m is a positive integer;
- W is the weight matrix of multiple feature matching algorithms;
- w i is the sorting weight corresponding to the i-th algorithm in multiple feature matching algorithms;
- FIRRi is the false identification rejection rate of the candidate feature template set corresponding to the i-th algorithm among multiple feature matching algorithms.
- the weighting module 403 is also used to: generate initial sorting matrices corresponding to multiple feature matching algorithms according to the corresponding candidate feature template sets under each feature matching algorithm; take each element of the initial sorting matrix as After the reciprocal, the reciprocal sorting matrix is generated; the reciprocal sorting matrix is standardized to obtain a standardized sorting matrix; the weight matrix is multiplied by the standardized sorting matrix to obtain the final sorting of the biometric samples corresponding to all candidate feature templates under multiple feature matching algorithms matrix.
- the determination module 404 is further configured to: arrange the elements in the final sorting matrix in descending order, and determine the candidate feature templates corresponding to the elements with the highest preset rank in the final sorting matrix as the biometric sample corresponding the final target object.
- biometric identification device 400 For other detailed descriptions of the biometric identification device 400, please refer to the descriptions of relevant method steps in the above embodiments.
- the embodiment of the present application also provides a device for biometric identification.
- the device may be a separate computer device, and the computer device may be a server, and its internal structure diagram may be shown in FIG. 5 .
- the computer device includes a processor, memory, network interface and database connected by a system bus. Wherein, the processor of the computer device is used to provide calculation and control capabilities.
- the memory of the computer device includes a non-volatile storage medium and an internal memory.
- the non-volatile storage medium stores an operating system, computer programs and databases.
- the internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium.
- the network interface of the computer device is used to communicate with an external terminal via a network connection.
- the device may also include an actual operating device using one or more of the methods described in this specification or one or more of the embodiments of the device.
- the biometric identification device may include at least one processor and a memory storing computer-executable instructions. When the processor executes the instructions, the steps of the biometric identification method in any one or more embodiments above are implemented.
- the embodiment of the present application also provides a non-transitory electronic device-readable storage medium, including: a program, which, when running on the electronic device, enables the electronic device to execute all or part of the procedures of the methods in the foregoing embodiments.
- the storage medium can be a magnetic disk, an optical disk, a read-only memory (Read-Only Memory, ROM), a random access memory (Random Access Memory, RAM), a flash memory (Flash Memory), a hard disk (Hard Disk Drive, Abbreviation: HDD) or Solid-State Drive (SSD), etc.
- the storage medium may also include a combination of the above-mentioned kinds of memories.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Human Computer Interaction (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Collating Specific Patterns (AREA)
Abstract
本申请提供一种生物特征识别方法、装置、设备和存储介质，该方法包括：获取待识别的生物特征样本；采用多种特征匹配算法，将所述生物特征样本与特征模板库中的特征模板进行逐一比对，得到每种所述特征匹配算法下对应的候选特征模板集合；根据权重矩阵中每种所述特征匹配算法的排序权重分别对每个所述候选特征模板集合的排序做加权计算；根据加权排序后的每个所述候选特征模板集合，确定所述生物特征样本对应的最终目标对象。本申请通过多种匹配算法的比对排名和排序权重，提高了生物特征辨识的可靠性。
Description
相关申请
本申请要求2021年11月15日申请的，申请号为202111350756.5，名称为“生物特征识别方法、装置、设备和存储介质”的中国专利申请的优先权，在此将其全文引入作为参考。
本申请涉及信息处理技术领域，尤其涉及一种生物特征识别方法、装置、设备和存储介质。
生物识别技术通过计算机与光学、声学、生物传感器和生物统计学原理等高科技手段密切结合，利用人体固有的生理特性如指纹、人脸、虹膜、指静脉等进行个人身份识别。
随着对社会安全和身份鉴别准确性和可靠性要求的日益提升，单一生物特征识别在准确性和可靠性方面的局限性日益突出，不能满足产品和技术发展的需要。随着生物特征识别技术的不断成熟，多模态生物特征识别技术的研究与应用将弥补单一生物特征识别技术的局限与不足，进一步降低生物特征识别系统的误识率，提高鉴别精度。
因此，多模态生物特征识别技术越来越受到人们的关注，生物特征辨识属于1:N的比对方式，在考勤门禁、人员聚档、黑名单监控等生物特征搜索中应用较多，给人们的生活和工作带来了便利。尽管如此，在1:N比对的实际应用中，往往会出现误识、误警及漏报的问题，给系统应用带来困扰和风险。
发明内容
本申请实施例的目的在于提供一种生物特征识别方法、装置、设备和存储介质，实现了通过多种匹配算法的比对排名和每种算法的排序权重来提高生物特征辨识的可靠性。
本申请实施例第一方面提供了一种生物特征识别方法，包括：获取待识别的生物特征样本；采用多种特征匹配算法，将所述生物特征样本与特征模板库中的特征模板进行逐一比对，得到每种所述特征匹配算法下对应的候选特征模板集合；根据权重矩阵中每种所述特征匹配算法的排序权重分别对每个所述候选特征模板集合的排序做加权计算；根据加权 排序后的每个所述候选特征模板集合，确定所述生物特征样本对应的最终目标对象。
在其中一个实施例中，所述采用多种特征匹配算法，将所述生物特征样本与特征模板库中的特征模板进行逐一比对，得到每种所述特征匹配算法下对应的候选特征模板集合，包括：采用所述多种特征匹配算法中的第一个特征匹配算法，将所述生物特征样本与所述特征模板库中的特征模板进行逐一比对，选取出第一候选特征模板集合；分别采用所述多种特征匹配算法中剩余的每种特征匹配算法，将所述生物特征样本与所述第一候选特征模板集合中的特征模板进行逐一比对，得到所述每种特征匹配算法下对应的第二候选特征模板集合。
在其中一个实施例中，在所述根据所述权重矩阵中每种所述特征匹配算法的排序权重分别对每个所述候选特征模板集合的排序做加权计算之前，所述生物特征识别方法还包括：分别根据每个所述候选特征模板集合的辨识错误率，构建所述多种特征匹配算法的所述权重矩阵，所述权重矩阵中包括每种所述特征匹配算法的排序权重。
在其中一个实施例中，所述辨识错误率包括：每个所述候选特征模板集合的错误辨识选择率；采用如下公式构建所述多种特征匹配算法的权重矩阵：
W＝(w
i)
1×m,i＝1～m
其中，m为所述多种特征匹配算法的总数量，m为正整数，W为所述多种特征匹配算法的权重矩阵，w
i为所述多种特征匹配算法中第i个算法对应的所述排序权重，FISRi为所述多种特征匹配算法中第i个算法对应的所述候选特征模板集合的错误辨识选择率。
在其中一个实施例中，所述辨识错误率包括：每个所述候选特征模板集合的错误辨识选择率和错误辨识拒绝率；采用如下公式构建所述多种特征匹配算法的权重矩阵：
W＝(w
i),i＝1～m
其中，m为所述多种特征匹配算法的总数量，m为正整数，W为所述多种特征匹配算法的权重矩阵，w
i为所述多种特征匹配算法中第i个算法对应的所述排序权重，FISRi为所述多种特征匹配算法中第i个算法对应的所述候选特征模板集合的错误辨识选择率，FIRRi为所述多种特征匹配算法中第i个算法对应的所述候选特征模板集合的错误辨识拒绝率。
在其中一个实施例中，所述根据所述权重矩阵中每种所述特征匹配算法的排序权重分别对每个所述候选特征模板集合的排序做加权计算，包括：根据每种所述特征匹配算法下对应的候选特征模板集合，生成所述多种特征匹配算法对应的初始排序矩阵；将所述初始 排序矩阵的每个元素取倒数后，生成倒数排序矩阵；对所述倒数排序矩阵进行标准化处理，得到标准化排序矩阵；将所述权重矩阵与所述标准化排序矩阵相乘，得到所述生物特征样本在所述多种特征匹配算法下对应所有候选特征模板的最终排序矩阵。
在其中一个实施例中，所述根据加权排序后的每个所述候选特征模板集合，确定所述生物特征样本对应的最终目标对象，包括：将所述最终排序矩阵中的元素按照降序排列，将所述最终排序矩阵中排在前预设名次的元素对应的候选特征模板确定为所述生物特征样本对应的最终目标对象。
本申请实施例第二方面提供了一种生物特征识别装置，包括：获取模块，用于获取待识别的生物特征样本；比对模块，用于采用多种特征匹配算法，将所述生物特征样本与特征模板库中的特征模板进行逐一比对，得到每种所述特征匹配算法下对应的候选特征模板集合；加权模块，用于根据权重矩阵中每种所述特征匹配算法的排序权重分别对每个所述候选特征模板集合的排序做加权计算；确定模块，用于根据加权排序后的每个所述候选特征模板集合，确定所述生物特征样本对应的最终目标对象。
在其中一个实施例中，所述比对模块用于：采用所述多种特征匹配算法中的第一个特征匹配算法，将所述生物特征样本与所述特征模板库中的特征模板进行逐一比对，选取出第一候选特征模板集合；分别采用所述多种特征匹配算法中剩余的每种特征匹配算法，将所述生物特征样本与所述第一候选特征模板集合中的特征模板进行逐一比对，得到所述每种特征匹配算法下对应的第二候选特征模板集合。
在其中一个实施例中，所述生物特征识别装置还包括：构建模块，用于在所述根据权重矩阵中每种所述特征匹配算法的排序权重分别对每个所述候选特征模板集合的排序做加权计算之前，分别根据每个所述候选特征模板集合的辨识错误率，构建所述多种特征匹配算法的所述权重矩阵，所述权重矩阵中包括每种所述特征匹配算法的排序权重。
在其中一个实施例中，所述辨识错误率包括：每个所述候选特征模板集合的错误辨识选择率；所述构建模块用于采用如下公式构建所述多种特征匹配算法的权重矩阵：
W＝(w
i)
1×m,i＝1～m
其中，m为所述多种特征匹配算法的总数量，m为正整数，W为所述多种特征匹配算法的权重矩阵，w
i为所述多种特征匹配算法中第i个算法对应的所述排序权重，FISRi为所述多种特征匹配算法中第i个算法对应的所述候选特征模板集合的错误辨识选择率。
在其中一个实施例中，所述辨识错误率包括：每个所述候选特征模板集合的错误辨识 选择率和错误辨识拒绝率；所述构建模块用于采用如下公式构建所述多种特征匹配算法的权重矩阵：
W＝(w
i),i＝1～m
其中，m为所述多种特征匹配算法的总数量，m为正整数，W为所述多种特征匹配算法的权重矩阵，w
i为所述多种特征匹配算法中第i个算法对应的所述排序权重，FISRi为所述多种特征匹配算法中第i个算法对应的所述候选特征模板集合的错误辨识选择率，FIRRi为所述多种特征匹配算法中第i个算法对应的所述候选特征模板集合的错误辨识拒绝率。
在其中一个实施例中，所述加权模块用于：根据每种所述特征匹配算法下对应的候选特征模板集合，生成所述多种特征匹配算法对应的初始排序矩阵；将所述初始排序矩阵的每个元素取倒数后，生成倒数排序矩阵；对所述倒数排序矩阵进行标准化处理，得到标准化排序矩阵；将所述权重矩阵与所述标准化排序矩阵相乘，得到所述生物特征样本在所述多种特征匹配算法下对应所有候选特征模板的最终排序矩阵。
在其中一个实施例中，所述确定模块用于：将所述最终排序矩阵中的元素按照降序排列，将所述最终排序矩阵中排在前预设名次的元素对应的候选特征模板确定为所述生物特征样本对应的最终目标对象。
本申请实施例第三方面提供了一种电子设备，包括：存储器，用于存储计算机程序；处理器，用于执行所述计算机程序，以实现本申请实施例第一方面中其任一实施例的方法。
本申请实施例第四方面提供了一种非暂态电子设备可读存储介质，包括：程序，当其藉由电子设备运行时，使得所述电子设备执行本申请实施例第一方面中其任一实施例的方法。
本申请提供的生物特征识别方法、装置、设备和存储介质，通过多种特征匹配算法，将待识别生物特征样本与特征模板库中的特征模板进行逐一比对，从而得到每种特征匹配算法下对应的候选特征模板集合，并根据每个候选特征模板集合下的辨识错误率构建多种特征匹配算法的权重矩阵，最后根据权重矩阵中的每种特征匹配算法的排序权重，分别对每个候选特征模板集合的排序做加权计算，如此，既利用了多种匹配算法的比对排名，又考虑了每种算法的排序权重，从而提高了生物特征辨识的可靠性。
为了更清楚地说明本申请实施例或传统技术中的技术方案，下面将对实施例或传统技 术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请的实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据公开的附图获得其他的附图。
图1为根据本申请一实施例的电子设备的结构示意图。
图2为根据本申请一实施例的生物特征识别方法的流程示意图。
图3为根据本申请一实施例的生物特征识别方法的流程示意图。
图4为根据本申请一实施例的生物特征识别装置的结构示意图。
图5为根据本申请一实施例的用于生物特征识别的设备内部结构示意图。
为使本申请的上述目的、特征和优点能够更加明显易懂，下面结合附图对本申请的具体实施方式做详细的说明。在下面的描述中阐述了很多具体细节以便于充分理解本申请。但是本申请能够以很多不同于在此描述的其它方式来实施，本领域技术人员可以在不违背本申请内涵的情况下做类似改进，因此本申请不受下面公开的具体实施例的限制。
在本申请的描述中，术语“第一”、“第二”等仅用于区分描述，而不能理解为指示或暗示相对重要性。
如图1所示，本实施例提供一种电子设备1，包括：至少一个处理器11和存储器12。图1中以一个处理器为例。处理器11和存储器12通过总线10连接。存储器12存储有可被处理器11执行的计算机指令，该计算机指令被处理器11执行，以使电子设备1可执行下述的实施例中方法的全部或部分流程，从而通过多种匹配算法的比对排名和每种算法的排序权重，提高生物特征辨识的可靠性。
在本申请一些实施例中，电子设备1可以是手机、平板电脑、笔记本电脑、台式计算机或者是由多个计算机等组成的服务器。
请参看图2，其为本申请一实施例的生物特征识别方法，该方法可由图1所示的电子设备1来执行，其通过多种匹配算法的比对排名和算法的排序权重，提高生物特征辨识的可靠性。该方法包括如下步骤：
步骤210：获取待识别的生物特征样本。
在本步骤中，待识别的生物特征样本指的是人体固有的生理特征如指纹、人脸、虹膜、指静脉等。
在本申请一些实施例中，若生物特征样本为指纹，指纹图像的获取方式可以为光学指纹采集技术、电容式传感器指纹采集技术、温度传感指纹获取技术、超声波指纹采集技术、 电磁波指纹采集技术等。
在本申请一些实施例中，使用图像采集软件可获待识别的指纹特征样本，如Anguli、ABIS多模态生物识别统一平台等。
步骤220：采用多种特征匹配算法，将生物特征样本与特征模板库中的特征模板进行逐一比对，得到每种特征匹配算法对应的候选特征模板集合。
在本步骤中，特征匹配算法指的是生物特征样本特征与特征模板库中的特征模板进行匹配的算法。
在本申请一些实施例中，若生物特征样本是人脸，人脸识别算法包括基于几何特征的人脸识别方法、基于子空间的人脸识别方法、基于统计的人脸识别方法、弹性图匹配方法、基于混合模型的人脸识别方法、基于三维模型的人脸识别方法以及基于深度神经网络的人脸识别方法等等。其中，基于统计的人脸识别方法又包括基于KL变换的人脸识别方法、基于隐马尔可夫模型的人脸识别方法等。
在本申请一些实施例中，多种特征匹配算法可以是不同类型的算法，比如第一种是基于深度神经网络的人脸识别方法，第二种是基于子空间的人脸识别方法，第三种又可以是不同类型的算法。具体地：
首先，将基于深度神经网络的人脸识别方法作为第一个特征匹配算法，使用基于深度神经网络的人脸识别方法将待识别人脸与人脸特征模板进行逐一比对，得到匹配分数，并将得到的匹配分数按照降序的方式进行排序；同时，将排序前k位对应的特征模板作为第一候选特征模板集合。其中，k为自定义的长度。
然后，使用基于子空间的人脸识别方法将待识别的人脸与第一候选特征模板集合中的特征模板逐一比对，得到匹配分数，并将得到的匹配分数按照降序的方式进行排序，从而得到基于子空间的人脸识别方法下的待识别的人脸对应的第二候选特征模板集合。
依次类推，分别采用剩余的人脸识别算法，将待识别的人脸样本与第一候选特征模板集合中的特征模板进行逐一比对，得到每种特征匹配算法对应的第二候选特征模板集合。
在本申请一些实施例中，多种特征匹配算法也可以同类型的算法，比如都是基于神经网络的算法，但是神经网络不同，特征匹配算法也会有差别。
步骤230：分别根据每个候选特征模板集合的辨识错误率，构建多种特征匹配算法的权重矩阵。
在本步骤中，权重矩阵中包括每种特征匹配算法的排序权重，可以预先统计确定每一种特征匹配算法的排序权重，以此构成权重矩阵。特征模板集合指的是注册集，即用于生物特征辨识(1:N比对)的注册模板集合。生物特征辨识算法性能体现在如错误辨识选择 率(FISR)、错误辨识拒绝率(FIRR)、首选识别率(TOP1)、前k选识别率(TOP k)等指标上。算法性能指标的评估在一定的测试样本集上进行。测试样本集可分为注册集和探针集；其中，探针集用于算法性能评测的待识别样本(也称为探针)的集合。
待识别样本在注册集中可能存在匹配的模板，也可能不存在。注册集和探针集内的样本数量可一致，也可以不一致。在生物特征辨识算法性能评测中，一般根据实际应用的数据规模来构建测试集的样本数量，如：万级、十万级、百万级、千万级等。算法性能指标的好坏与测试样本质量有关。样本集内的测试样本质量应遵循生物特征模态对应的样本质量标准。
其中，辨识错误率包括错误辨识选择率(FISR)和错误辨识拒绝率(FIRR)。FISR指标与注册集样本数量N、候选列表长度K、比对阈值T等有关，可用FISR(N,K,T)表示。FISR指标的统计方法是在不可匹配目标人的搜索中，比对结果列表中有得分超过阈值T且排名在K位以内的候选者(都是错误的匹配者)，FISR(N,K,T)为候选者数量与探针集中不可匹配样本数量的比值。
计算方式如下：
式中：N为注册集样本数量、K为候选列表长度、T为比对阈值，C
2为C
1轮比对搜索中得分超过阈值T且排名在K位以内的候选者数量之和；C
1为待识别样本的集合中不可匹配样本数量。
在N、K一定的情况下，随着T的增大，FISR(N,K,T)减小，说明错误匹配的概率越低，辨识可靠性增加。
FIRR指标与注册集样本数量N、候选列表长度K、比对阈值等有关T，可用FIRR(N,K,T)表示。统计方法为在可匹配目标人的搜索中，在比对结果列表中存在探针与对应模板的比对得分未超过阈值T或排名在K位以外的被错误拒绝者(是应匹配而未匹配的失败者)，FIRR(N,K,T)为被错误拒绝的数量与探针集中可匹配样本数量的比值。
计算方式如下：
式中：C
4为C
3轮比对搜索中得分未超过阈值T或排名在K位以外的被错误拒绝的数量；C
3为待识别样本的集合中可匹配样本数量。
在FISR(N，K，T)一定的情况下，FIRR(N，K，T)越小，说明被拒绝的概率越小，相当于识别通过率增加，因此辨识可靠性增加。
通常，在N、K一定的前提下，调整阈值T，使FISR(N，K，T)达到可用的可靠性等级(如：万分之一、十万分之一)，根据此时的阈值T，就可以确定FIRR(N，K，T)的值。
在本申请一些实施例中，采用如下公式构建多种特征匹配算法的权重矩阵：
W＝(w
i)
1×m,i＝1～m
式中：m为多种特征匹配算法的总数量，m为正整数；W为多种特征匹配算法的权重矩阵；w
i为多种特征匹配算法中第i个算法对应的排序权重；FISRi为多种特征匹配算法中第i个算法对应的候选特征模板集合的错误辨识选择率。
在本申请一些实施例中，采用如下公式构建多种特征匹配算法的权重矩阵：
W＝(w
i),i＝1～m
式中：m为多种特征匹配算法的总数量，m为正整数；W为多种特征匹配算法的权重矩阵；w
i为多种特征匹配算法中第i个算法对应的排序权重；FISRi为多种特征匹配算法中第i个算法对应的候选特征模板集合的错误辨识选择率；FIRRi为多种特征匹配算法中第i个算法对应的候选特征模板集合的错误辨识拒绝率。
需要说明的是，步骤230中对权重矩阵的构建需要发生在步骤240之前，在使用的多种特征匹配算法确定后，权重矩阵也就可以确定。在实际进行生物特征识别时，只要特征匹配算法不变，就无需每次构建权重矩阵，因此步骤230也可以发生在步骤210-步骤220之前，或者同时进行。本实施例对步骤230与步骤210、步骤220之间的发生次序不做限定。
步骤240：根据所述权重矩阵中每种特征匹配算法的排序权重分别对每个候选特征模板集合的排序做加权计算。
在本步骤中，每个候选特征模板集合的排序指的是每个候选特征模板的排序分数组成的矩阵。
步骤250：根据加权排序后的每个候选特征模板集合，确定生物特征样本对应的最终目标对象。
在本步骤中，最终目标对象可以是加权排序后排名第一的目标对象，也可以是按照设定的长度进行降序返回的候选者的结果集合。
请参看图3其为本申请一实施例的生物特征识别方法，该方法包括如下步骤：
步骤310：获取待识别的生物特征样本。详细参见上述实施例中对步骤210的描述。
步骤320：采用多种特征匹配算法，将生物特征样本与特征模板库中的特征模板进行逐一比对，得到每种特征匹配算法下对应的候选特征模板集合。详细参见上述实施例中对步骤220的描述。
步骤330：分别根据每个候选特征模板集合的辨识错误率，构建多种特征匹配算法的权重矩阵。详细参见上述实施例中对步骤230的描述。
步骤340：根据每种特征匹配算法下对应的候选特征模板集合，生成多种特征匹配算法对应的初始排序矩阵。
步骤350：将初始排序矩阵的每个元素取倒数后，生成倒数排序矩阵。
在本步骤中，初始排序矩阵指的是根据每轮比对候选列表后生成排序矩阵(r
ij)
m×k，具体形式如下：
其中：m为待识别生物特征样本经过了m轮的比对，k为候选集合的长度，r
ij为待识别生物特征样本在i轮的匹配后排名为j的匹配分数。
将排序矩阵每个元素取倒数，即将(r
ij)
m×k进行矩阵变换，对r
ij取倒数可得：b
ij＝1/r
ij；
从而得到倒数排序矩阵，具体形式如下：
步骤360：对倒数排序矩阵进行标准化处理，得到标准化排序矩阵。
在本步骤中，标准化处理指的是将倒数排序矩阵中的元素进行z-分数变换，从而得到标准化排序矩阵(z
ij)
m×k，具体形式如下：
其中，z-分数变换方式如下：
z
ij＝(b
ij-μ)/δ；
式中：b
ij为倒数排序矩阵中第i行第j列数值，μ为b
ij的平均值，δ为b
ij的标准差。
步骤370：将权重矩阵与标准化排序矩阵相乘，得到生物特征样本在多种特征匹配算法下对应所有候选特征模板的最终排序矩阵。
在本步骤中，权重矩阵与标准化排序矩阵相乘的具体公式如下：
(c
j)
1×k＝(w
i)
1×m*(z
ij)
m×k；
则加权后的排序矩阵为：C＝(c
1,c
2,…,c
k),并对其矩阵内的元素进行降序排列，从而得到最终排序矩阵。
步骤380：根据加权排序后的每个候选特征模板集合，确定生物特征样本对应的最终目标对象。详细参见上述实施例中对步骤250的描述。
请参看图4，其为本申请一实施例提供的生物特征识别装置400，该装置可应用于图1所示的电子设备1，其通过利用多种匹配算法的比对排名和每种算法的排序权重，提高了生物特征辨识的可靠性。该装置包括：获取模块401、比对模块402、加权模块403和确定模块404，各个模块的原理关系如下：
获取模块401，用于获取待识别的生物特征样本。
比对模块402，用于采用多种特征匹配算法，将生物特征样本与特征模板库中的特征模板进行逐一比对，得到每种特征匹配算法下对应的候选特征模板集合；
加权模块403，用于根据权重矩阵中每种特征匹配算法的排序权重分别对每个候选特征模板集合的排序做加权计算；
确定模块404，用于根据加权排序后的每个候选特征模板集合，确定生物特征样本对应的最终目标对象。
在本申请一些实施例中，比对模块402还用于：采用多种特征匹配算法中的第一个特征匹配算法，将生物特征样本与特征模板库中的特征模板进行逐一比对，选取出第一候选特征模板集合；分别采用多种特征匹配算法中剩余的每种特征匹配算法，将生物特征样本与第一候选特征模板集合中的特征模板进行逐一比对，得到每种特征匹配算法下对应的第二候选特征模板集合。
在本申请一些实施例中，该生物特征识别装置400还包括：构建模块405，用于在根据权重矩阵中每种特征匹配算法的排序权重分别对每个候选特征模板集合的排序做加权计算之前，分别根据每个候选特征模板集合的辨识错误率，构建多种特征匹配算法的权重矩阵，其中权重矩阵包括每种特征匹配算法的排序权重。
在本申请一些实施例中，辨识错误率包括：每个候选特征模板集合的错误辨识选择率；构建模块405还用于采用如下公式构建多种特征匹配算法的权重矩阵：
W＝(w
i)
1×m,i＝1～m
其中，m为多种特征匹配算法的总数量，m为正整数；W为多种特征匹配算法的权重矩阵；w
i为多种特征匹配算法中第i个算法对应的排序权重；FISRi为多种特征匹配算法中第i个算法对应的候选特征模板集合的错误辨识选择率。
在本申请一些实施例中，辨识错误率包括：每个候选特征模板集合的错误辨识选择率和错误辨识拒绝率；构建模块405还用于采用如下公式构建多种特征匹配算法的权重矩阵：
W＝(w
i),i＝1～m
其中，m为多种特征匹配算法的总数量，m为正整数；W为多种特征匹配算法的权重矩阵；w
i为多种特征匹配算法中第i个算法对应的排序权重；FISRi为多种特征匹配算法中第i个算法对应的候选特征模板集合的错误辨识选择率；FIRRi为多种特征匹配算法中第i个算法对应的候选特征模板集合的错误辨识拒绝率。
在本申请一些实施例中，加权模块403还用于：根据每种特征匹配算法下对应的候选特征模板集合，生成多种特征匹配算法对应的初始排序矩阵；将初始排序矩阵的每个元素取倒数后，生成倒数排序矩阵；对倒数排序矩阵进行标准化处理，得到标准化排序矩阵；将权重矩阵与标准化排序矩阵相乘，得到生物特征样本在多种特征匹配算法下对应所有候选特征模板的最终排序矩阵。
在本申请一些实施例中，确定模块404还用于：将最终排序矩阵中的元素按照降序排列，将最终排序矩阵中排在前预设名次的元素对应的候选特征模板确定为生物特征样本对应的最终目标对象。
上述生物特征识别装置400的其他详细描述，请参见上述实施例中相关方法步骤的描述。
本申请实施例还提供了一种用于生物特征识别的设备，所述的设备可以为单独的计算机设备，该计算机设备可以是服务器，其内部结构图可以如图5所示。该计算机设备包括通过系统总线连接的处理器、存储器、网络接口和数据库。其中，该计算机设备的处理器用于提供计算和控制能力。该计算机设备的存储器包括非易失性存储介质、内存储器。该非易失性存储介质存储有操作系统、计算机程序和数据库。该内存储器为非易失性存储介质中的操作系统和计算机程序的运行提供环境。该计算机设备的网络接口用于与外部的终端通过网络连接通信。该计算机程序被处理器执行时以实现一种生物特征识别方法。该设备也可以包括使用了本说明书的一个或多个所述方法或一个或多个实施例装置的实际操作装置等。所述生物特征识别设备可以包括至少一个处理器以及存储计算机可执行指令的 存储器，处理器执行所述指令时实现上述任意一个或者多个实施例中所述生物特征识别方法的步骤。
本申请实施例还提供了一种非暂态电子设备可读存储介质，包括：程序，当其在电子设备上运行时，使得电子设备可执行上述实施例中方法的全部或部分流程。其中，存储介质可为磁盘、光盘、只读存储记忆体(Read-Only Memory，ROM)、随机存储记忆体(Random Access Memory，RAM)、快闪存储器(Flash Memory)、硬盘(Hard Disk Drive，缩写：HDD)或固态硬盘(Solid-State Drive，SSD)等。存储介质还可以包括上述种类的存储器的组合。
以上实施例的各技术特征可以进行任意的组合，为使描述简洁，未对上述实施例中的各个技术特征所有可能的组合都进行描述，然而，只要这些技术特征的组合不存在矛盾，都应当认为是本说明书记载的范围。
以上所述实施例仅表达了本申请的几种实施方式，其描述较为具体和详细，但并不能因此而理解为对发明专利范围的限制。应当指出的是，对于本领域的普通技术人员来说，在不脱离本申请构思的前提下，还可以做出若干变形和改进，这些都属于本申请的保护范围。因此，本申请专利的保护范围应以所附权利要求为准。
Claims (15)
- 一种生物特征识别方法，包括：获取待识别的生物特征样本；采用多种特征匹配算法，将所述生物特征样本与特征模板库中的特征模板进行逐一比对，得到每种所述特征匹配算法下对应的候选特征模板集合；根据权重矩阵中每种所述特征匹配算法的排序权重分别对每个所述候选特征模板集合的排序做加权计算；根据加权排序后的每个所述候选特征模板集合，确定所述生物特征样本对应的最终目标对象。
- 根据权利要求1所述的方法，其中，所述采用多种特征匹配算法，将所述生物特征样本与特征模板库中的特征模板进行逐一比对，得到每种所述特征匹配算法下对应的候选特征模板集合，包括：采用所述多种特征匹配算法中的第一个特征匹配算法，将所述生物特征样本与所述特征模板库中的特征模板进行逐一比对，选取出第一候选特征模板集合；分别采用所述多种特征匹配算法中剩余的每种特征匹配算法，将所述生物特征样本与所述第一候选特征模板集合中的特征模板进行逐一比对，得到所述每种特征匹配算法下对应的第二候选特征模板集合。
- 根据权利要求1所述的方法，其中，在所述根据权重矩阵中每种所述特征匹配算法的排序权重分别对每个所述候选特征模板集合的排序做加权计算之前，还包括：分别根据每个所述候选特征模板集合的辨识错误率，构建所述多种特征匹配算法的所述权重矩阵，其中所述权重矩阵中包括每种所述特征匹配算法的排序权重。
- 根据权利要求3所述的方法，其中，所述辨识错误率包括：每个所述候选特征模板集合的错误辨识选择率；采用如下公式构建所述多种特征匹配算法的权重矩阵：W＝(w i) 1×m,i＝1～m
其中，m为所述多种特征匹配算法的总数量，m为正整数，W为所述多种特征匹配算法的权重矩阵，w i为所述多种特征匹配算法中第i个算法对应的所述排序权重，FISRi为所述多种特征匹配算法中第i个算法对应的所述候选特征模板集合的错误辨识选择率。 - 根据权利要求3所述的方法，其中，所述辨识错误率包括：每个所述候选特征模板 集合的错误辨识选择率和错误辨识拒绝率；采用如下公式构建所述多种特征匹配算法的权重矩阵：W＝(w i),i＝1～m
其中，m为所述多种特征匹配算法的总数量，m为正整数，W为所述多种特征匹配算法的权重矩阵，w i为所述多种特征匹配算法中第i个算法对应的所述排序权重，FISRi为所述多种特征匹配算法中第i个算法对应的所述候选特征模板集合的错误辨识选择率，FIRRi为所述多种特征匹配算法中第i个算法对应的所述候选特征模板集合的错误辨识拒绝率。 - 根据权利要求1所述的方法，其中，所述根据所述权重矩阵中每种所述特征匹配算法的排序权重分别对每个所述候选特征模板集合的排序做加权计算，包括：根据每种所述特征匹配算法下对应的候选特征模板集合，生成所述多种特征匹配算法对应的初始排序矩阵；将所述初始排序矩阵的每个元素取倒数后，生成倒数排序矩阵；对所述倒数排序矩阵进行标准化处理，得到标准化排序矩阵；将所述权重矩阵与所述标准化排序矩阵相乘，得到所述生物特征样本在所述多种特征匹配算法下对应所有候选特征模板的最终排序矩阵。
- 根据权利要求6所述的方法，其中，所述根据加权排序后的每个所述候选特征模板集合，确定所述生物特征样本对应的最终目标对象，包括：将所述最终排序矩阵中的元素按照降序排列，将所述最终排序矩阵中排在前预设名次的元素对应的候选特征模板确定为所述生物特征样本对应的最终目标对象。
- 根据权利要求6所述的方法，其中，所述初始排序矩阵指的是根据每轮比对候选列表后生成排序矩阵(r ij) m×k，具体形式如下：
其中：m为待识别生物特征样本经过了m轮的比对，k为候选集合的长度，r ij为待识别生物特征样本在i轮的匹配后排名为j的匹配分数。 - 根据权利要求6所述的方法，其中，标准化处理指的是将倒数排序矩阵中的元素进行z-分数变换，从而得到标准化排序矩阵(z ij) m×k，具体形式如下：
其中，z-分数变换方式如下：z ij＝(b ij-θ)/δ；其中b ij为倒数排序矩阵中第i行第j列数值，μ为b ij的平均值，δ为b ij的标准差。 - 一种生物特征识别装置，包括：获取模块，用于获取待识别的生物特征样本；比对模块，用于采用多种特征匹配算法，将所述生物特征样本与特征模板库中的特征模板进行逐一比对，得到每种所述特征匹配算法下对应的候选特征模板集合；加权模块，用于根据权重矩阵中每种所述特征匹配算法的排序权重分别对每个所述候选特征模板集合的排序做加权计算；确定模块，用于根据加权排序后的每个所述候选特征模板集合，确定所述生物特征样本对应的最终目标对象。
- 根据权利要求8所述的装置，还包括：构建模块，用于在所述根据所述权重矩阵中每种所述特征匹配算法的排序权重分别对每个所述候选特征模板集合的排序做加权计算之前，分别根据每个所述候选特征模板集合的辨识错误率，构建所述多种特征匹配算法的所述权重矩阵，其中所述权重矩阵中包括每种所述特征匹配算法的排序权重。
- 根据权利要求8所述的装置，其中，加权模块还用于：根据每种所述特征匹配算法下对应的候选特征模板集合，生成所述多种特征匹配算法对应的初始排序矩阵；将所述初始排序矩阵的每个元素取倒数后，生成倒数排序矩阵；对所述倒数排序矩阵进行标准化处理，得到标准化排序矩阵；将所述权重矩阵与所述标准化排序矩阵相乘，得到所述生物特征样本在所述多种特征匹配算法下对应所有候选特征模板的最终排序矩阵。
- 根据权利要求12所述的装置，其中所述初始排序矩阵指的是根据每轮比对候选列表后生成排序矩阵(r ij) m×k，具体形式如下：
其中：m为待识别生物特征样本经过了m轮的比对，k为候选集合的长度，r ij为待识别生物特征样本在i轮的匹配后排名为j的匹配分数。 - 一种电子设备，包括：存储器，用于存储计算机程序；处理器，用于执行所述计算机程序，以实现如权利要求1至9中任一项所述的方法。
- 一种计算机可读存储介质，包括用于存储处理器可执行指令的存储器，所述指令被所述处理器执行时实现权利要求1-9中任一项所述的方法。
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111350756.5A CN114049511A (zh) | 2021-11-15 | 2021-11-15 | 生物特征识别方法、装置、设备和存储介质 |
| CN202111350756.5 | 2021-11-15 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2023083051A1 true WO2023083051A1 (zh) | 2023-05-19 |
Family
ID=80209333
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2022/128883 WO2023083051A1 (zh) | 2021-11-15 | 2022-11-01 | 生物特征识别方法、装置、设备和存储介质 |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN114049511A (zh) |
| WO (1) | WO2023083051A1 (zh) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114049511A (zh) * | 2021-11-15 | 2022-02-15 | 北京眼神智能科技有限公司 | 生物特征识别方法、装置、设备和存储介质 |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150331867A1 (en) * | 2014-05-15 | 2015-11-19 | Bio-Key International, Inc. | Adaptive short lists and acceleration of biometric database search |
| CN110909757A (zh) * | 2019-08-20 | 2020-03-24 | 北京北科慧识科技股份有限公司 | 生物识别系统模板的选取和更新方法 |
| CN113204968A (zh) * | 2021-05-28 | 2021-08-03 | 平安科技（深圳）有限公司 | 医学实体的概念识别方法、装置、设备及存储介质 |
| CN114049511A (zh) * | 2021-11-15 | 2022-02-15 | 北京眼神智能科技有限公司 | 生物特征识别方法、装置、设备和存储介质 |
-
2021
- 2021-11-15 CN CN202111350756.5A patent/CN114049511A/zh active Pending
-
2022
- 2022-11-01 WO PCT/CN2022/128883 patent/WO2023083051A1/zh active Application Filing
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150331867A1 (en) * | 2014-05-15 | 2015-11-19 | Bio-Key International, Inc. | Adaptive short lists and acceleration of biometric database search |
| CN110909757A (zh) * | 2019-08-20 | 2020-03-24 | 北京北科慧识科技股份有限公司 | 生物识别系统模板的选取和更新方法 |
| CN113204968A (zh) * | 2021-05-28 | 2021-08-03 | 平安科技（深圳）有限公司 | 医学实体的概念识别方法、装置、设备及存储介质 |
| CN114049511A (zh) * | 2021-11-15 | 2022-02-15 | 北京眼神智能科技有限公司 | 生物特征识别方法、装置、设备和存储介质 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114049511A (zh) | 2022-02-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11282295B2 (en) | Image feature acquisition | |
| US9286528B2 (en) | Multi-modal biometric database searching methods | |
| US8031981B2 (en) | Method and systems for generating a subset of biometric representations | |
| Tao et al. | Fingerprint recognition with identical twin fingerprints | |
| CN109817339B (zh) | 基于大数据的患者分组方法和装置 | |
| US11710297B2 (en) | System and method for detecting potential matches between a candidate biometric and a dataset of biometrics | |
| Yue et al. | Hashing based fast palmprint identification for large-scale databases | |
| US20200387994A1 (en) | System and method for detecting potential fraud between a probe biometric and a dataset of biometrics | |
| US11062120B2 (en) | High speed reference point independent database filtering for fingerprint identification | |
| WO2021120587A1 (zh) | 基于oct的视网膜分类方法、装置、计算机设备及存储介质 | |
| Dong et al. | Open-set face identification with index-of-max hashing by learning | |
| WO2023083051A1 (zh) | 生物特征识别方法、装置、设备和存储介质 | |
| Soldaini et al. | Inferring individual attributes from search engine queries and auxiliary information | |
| CN112528022A (zh) | 主题类别对应的特征词提取和文本主题类别识别方法 | |
| CN110751171A (zh) | 图像数据分类方法、装置、计算机设备和存储介质 | |
| CN110414562A (zh) | X光片的分类方法、装置、终端及存储介质 | |
| Zhao | Learning user keystroke patterns for authentication | |
| US10198613B2 (en) | Latent fingerprint pattern estimation | |
| Indrawan et al. | Review of sequential access method for fingerprint identification | |
| Garain et al. | Addressing facial dynamics using k-medoids cohort selection algorithm for face recognition | |
| Bolourchi et al. | A machine learning-based data-driven approach to Alzheimer’s disease diagnosis using statistical and harmony search methods | |
| Di Martino et al. | A statistical approach to reliability estimation for fingerprint recognition | |
| CN111428679B (zh) | 影像识别方法、装置和设备 | |
| Aditya et al. | Face recognition using deep learning as user login on healthcare kiosk | |
| Tsai et al. | Blind Clustering of Fingerprints for Database Indexing. |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 22891847 Country of ref document: EP Kind code of ref document: A1 |
|
| WWE | Wipo information: entry into national phase |
Ref document number: 11202403159X Country of ref document: SG |