WO2019200902A1 - 图像识别方法及装置 - Google Patents
图像识别方法及装置 Download PDFInfo
- Publication number
- WO2019200902A1 WO2019200902A1 PCT/CN2018/115612 CN2018115612W WO2019200902A1 WO 2019200902 A1 WO2019200902 A1 WO 2019200902A1 CN 2018115612 W CN2018115612 W CN 2018115612W WO 2019200902 A1 WO2019200902 A1 WO 2019200902A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- face recognition
- recognition algorithm
- algorithm
- face
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Ceased
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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 field of face recognition technology, and in particular to an image recognition method and apparatus.
- Face recognition is a kind of biometric recognition technology based on human facial feature information. It belongs to the artificial intelligence biometric identification method which is widely used at present, also called face recognition and face recognition. With the development of artificial intelligence technology and computer recognition technology, face recognition technology has been widely used in various fields, and with the further maturity of technology and the improvement of social recognition, face recognition technology will be applied in more fields.
- the face recognition system is mainly optimized by the face recognition algorithm.
- recognition algorithms There are two existing recognition algorithms that have been widely used. One is to apply a single algorithm, but the single algorithm has certain limitations and is applied to specific In the application scenario, there is a problem that the scene migration causes the accuracy to decrease, and if the algorithm of a single specific scenario is applied to another specific scenario, there is a problem that the scene change causes the accuracy to decrease.
- the other is to provide a multi-algorithm face recognition fusion system, which obtains different modal pictures of the same face according to different modal cameras, and then each modal applies a specific algorithm for recognition.
- the drawback is that multimodal image acquisition will be more complicated, costly, and the similarity scales of different modalities are different, which is more difficult in the fractional fusion phase.
- At least some embodiments of the present application provide an image recognition method and apparatus to solve at least the technical problem of low recognition accuracy when using a plurality of recognition algorithms to recognize a face image.
- an image recognition method including: acquiring at least one face recognition algorithm; assigning corresponding weights to at least one face recognition algorithm, and determining at least one face Identifying a weight value of the algorithm; determining at least one face recognition algorithm by comparing at least one face recognition algorithm with the initial recognition result of the preset image and the preset recognition result; determining at least one face recognition algorithm according to at least one face recognition algorithm; The weight value and the similarity value are used to fuse the initial recognition result of at least one face recognition algorithm to obtain a recognition result of identifying the preset image.
- the pair is identified as the logarithm of the same person, TN is the logarithm of all negative sample pairs identified as different ones, NUM_P is the total number of positive sample pairs, and NUM_N is the total number of negative sample pairs; according to various face recognition algorithms
- the size of the test accuracy is determined to determine at least one face recognition algorithm.
- the method before acquiring the at least one face recognition algorithm, the method further includes: collecting a predetermined number of face image data in the current scene to obtain an image test set.
- determining at least one face recognition algorithm according to the size of the test accuracy of the plurality of face recognition algorithms including: performing multiple face recognition algorithms according to the test accuracy of the plurality of face recognition algorithms Sorting, selecting at least one face recognition algorithm from the sorting result; or determining at least one face recognition algorithm by comparing test accuracy of the plurality of face recognition algorithms with a preset threshold.
- At least one of the face recognition algorithms is assigned a corresponding weight by at least one of the following methods: for at least one face recognition algorithm, the weight value is 1; according to at least one face recognition algorithm Arranging the order, assigning weight to at least one face recognition algorithm; assigning weight to at least one face recognition algorithm according to the test accuracy of at least one face recognition algorithm.
- the initial recognition result of the at least one face recognition algorithm is merged according to the weight value and the similarity value of the at least one face recognition algorithm, and the recognition result of the preset image is obtained, including: at least one The similarity score vector of each face recognition algorithm in the face recognition algorithm is sorted to obtain a sort result; according to the sort result, a predetermined number of similarity score vectors in each face recognition algorithm are determined as similarity values; The degree value and the weight value are fused to the initial recognition result of the at least one face recognition algorithm to obtain a recognition result of identifying the preset image.
- performing a fusion process on the initial recognition result of the at least one face recognition algorithm according to the weight value and the similarity value of the at least one face recognition algorithm, to obtain the recognition result of identifying the preset image including: determining at least one The similarity threshold of each face recognition algorithm in the face recognition algorithm; comparing the similarity score vector and the similarity threshold of each face recognition algorithm; if the similarity score vector is greater than or equal to the similarity threshold, it will be similar The degree score vector is determined as the similarity value; according to the similarity value and the weight value, the initial recognition result of the at least one face recognition algorithm is merged to obtain the recognition result of the recognized preset image.
- an image recognition apparatus including: an acquisition module configured to acquire at least one face recognition algorithm; and an assignment module configured to respectively identify at least one face The algorithm assigns a corresponding weight to determine a weight value of the at least one face recognition algorithm; and the determining module is configured to determine at least one of the initial recognition result of the preset image and the preset recognition result by comparing at least one face recognition algorithm a similarity value of the face recognition algorithm; the identification module is configured to fuse the initial recognition result of the at least one face recognition algorithm according to the weight value and the similarity value of the at least one face recognition algorithm to obtain the recognition preset The recognition result of the image.
- a storage medium comprising a stored program, wherein the program performs any of the optional or preferred image recognition methods described above.
- processor being arranged to execute a program, wherein the program is executed to perform any of the optional or preferred image recognition methods described above.
- At least one face recognition algorithm is obtained by acquiring at least one face recognition algorithm, and weights of at least one face recognition algorithm are determined by comparing at least one of the face recognition algorithms;
- the face recognition algorithm identifies an initial recognition result of the preset image and a preset recognition result, and determines a similarity value of the at least one face recognition algorithm; and at least one according to the weight value and the similarity value of the at least one face recognition algorithm.
- the initial recognition result of the face recognition algorithm is fused, and the recognition result of the preset image is obtained, which achieves the purpose of improving the recognition accuracy of the universal recognition algorithm, thereby realizing the application of the algorithm of multiple common scenes to a specific scene.
- the technical effect further solves the technical problem that the recognition accuracy is low when a plurality of recognition algorithms are used to recognize the face image.
- FIG. 1 is a flow chart showing the steps of an image recognition method according to an embodiment of the present application.
- FIG. 2 is a flow chart showing the steps of an optional image recognition method according to an embodiment of the present application.
- FIG. 3 is a flow chart showing the steps of an optional image recognition method according to an embodiment of the present application.
- FIG. 4 is a flow chart showing the steps of an optional image recognition method according to an embodiment of the present application.
- FIG. 5 is a schematic structural diagram of an image recognition apparatus according to an embodiment of the present application.
- an embodiment of an image recognition method is provided, it being noted that the steps illustrated in the flowchart of the drawings may be performed in a computer system such as a set of computer executable instructions, and, although The logical order is shown in the flowchart, but in some cases the steps shown or described may be performed in a different order than the ones described herein.
- the application can apply a plurality of different face recognition algorithms to a specific scene for face recognition, and has a wide application scenario and strong adaptability, and is applied to a specific scene through a recognition algorithm of several common scenes. And can improve the recognition accuracy of the universal recognition algorithm.
- FIG. 1 is a flow chart of steps of an image recognition method according to an embodiment of the present application. As shown in FIG. 1, the method includes the following steps:
- Step S102 Acquire at least one face recognition algorithm.
- step S102 if there are multiple face recognition algorithms in a single face recognition scenario, multiple face recognition algorithms need to be sorted and filtered in the current scene to obtain at least one face recognition. algorithm.
- an image test set (eg, a face image test set) of the current scene may be tested by multiple face recognition algorithms, and the face recognition algorithm is used on the image test set according to each face recognition algorithm. Test accuracy, select at least one face recognition algorithm in descending order.
- Step S104 assign weights to at least one face recognition algorithm to determine weight values of at least one face recognition algorithm.
- each face recognition algorithm in the selected at least one face recognition algorithm may be weighted in different manners. If n face recognition algorithms are selected, The weight values assigned by the n face recognition algorithms may be [W1, W2, W3...Wn], where W is a weight value, and 1, 2, 3...n are 1-n face recognition algorithms.
- At least one of the face recognition algorithms is assigned a corresponding weight by at least one of the following manners:
- Manner 2 assigning weight to at least one face recognition algorithm according to an arrangement order of at least one face recognition algorithm
- Manner 3 At least one face recognition algorithm is weighted according to the test accuracy of at least one face recognition algorithm.
- the ratio of the accuracy of the selected n face recognition algorithms to the weight of the weights is obtained, thereby determining n
- Step S106 Identify, by comparing at least one face recognition algorithm, an initial recognition result of the preset image and a preset recognition result, and determine a similarity value of the at least one face recognition algorithm;
- Step S108 Perform fusion processing on the initial recognition result of the at least one face recognition algorithm according to the weight value and the similarity value of the at least one face recognition algorithm, to obtain a recognition result of identifying the preset image.
- the initial recognition result is that each face recognition algorithm in the at least one face recognition algorithm recognizes the recognition result of the preset image, and the preset recognition result is a preset correct recognition result.
- a face image database may be pre-established, and the face database may be used to store a plurality of personnel information (where the person information may include a person's identity) and a plurality of test pictures.
- the face database may be preset that the face database has P personnel information, and then each of the selected face recognition algorithms in the selected n face recognition algorithms is used to test a certain test photo, and P similarity values may be obtained.
- P similarity values can be any value between 0-1.
- the i-th face recognition algorithm identifies the initial recognition result of the test photo, which may be [(Si1, identity 1), (Si2, identity 2), (Si3, identity 3)... (SiP, identity P )].
- the initial recognition result of the at least one face recognition algorithm may be fused according to the P similarity values and the weight values of the n types of face recognition algorithms, to obtain the recognized preset image. Identification result.
- the identity of a person with the highest similarity vector may be determined by sorting the similarity vector in the final recognition result, which is the recognition result of the test picture.
- the fusion processing can be implemented by using a plurality of different face recognition algorithms to obtain the recognition result, and the recognition accuracy of the plurality of recognition algorithms in the current specific scene is improved.
- At least one face recognition algorithm is obtained, and weights of at least one face recognition algorithm are determined by assigning corresponding weights to at least one face recognition algorithm; by comparing at least one face
- the recognition algorithm identifies an initial recognition result of the preset image and a preset recognition result, and determines a similarity value of the at least one face recognition algorithm; and at least one person according to the weight value and the similarity value of the at least one face recognition algorithm
- the initial recognition result of the face recognition algorithm is fused, and the recognition result of the preset image is obtained, which achieves the purpose of improving the recognition accuracy of the universal recognition algorithm, thereby realizing the technology of applying the fusion of multiple common scene algorithms to a specific scene.
- the effect further solves the technical problem that the recognition accuracy is low when a plurality of recognition algorithms are used to identify the face image.
- FIG. 2 is a flowchart of steps of an optional image recognition method according to an embodiment of the present application. As shown in FIG. 2, at least one face recognition algorithm is acquired, including the following steps. :
- Step S204 Determine at least one face recognition algorithm according to the size of the test accuracy of the plurality of face recognition algorithms.
- the method before acquiring at least one face recognition algorithm, the method further includes the following steps: collecting a predetermined number of face image data in the current scene to obtain an image test set.
- the foregoing face image data may be collected by taking a photo.
- the face image data of the current application scenario may be collected.
- the current application scenario may have 100 A person, 5 photos per person, a total of 500 pictures, and then a face image test set can be constructed, wherein the face image test set includes: 1000 pairs of positive sample pairs and negative sample pairs.
- the positive sample pair and the negative sample in the above face image test set can be obtained by:
- determining at least one face recognition algorithm according to the size of the test accuracy of the plurality of face recognition algorithms includes the following steps:
- the plurality of face recognition algorithms are sorted, and at least one face recognition algorithm is selected from the sorting results;
- At least one face recognition algorithm is determined by comparing the test accuracy of the plurality of face recognition algorithms with a preset threshold.
- the pre-sorting n personal face recognition algorithm may be directly selected according to the test accuracy of the plurality of face recognition algorithms.
- a test accuracy threshold may be given in advance. If there are K face recognition algorithms whose test accuracy is greater than or equal to the test accuracy threshold, if K is greater than or equal to n, then The first n face recognition algorithm is the above-mentioned at least one face recognition algorithm. If K is smaller than n, the K face recognition algorithm is selected as the at least one face recognition algorithm.
- FIG. 3 is a flow chart of steps of an optional image recognition method according to an embodiment of the present application, as shown in FIG. 3, according to weight values of at least one face recognition algorithm.
- the similarity value is obtained by fusing the initial recognition result of the at least one face recognition algorithm, and obtaining the recognition result of identifying the preset image, comprising the following steps:
- Step S302 sorting the similarity score vectors of each face recognition algorithm in at least one face recognition algorithm to obtain a sort result
- Step S304 determining, according to the sorting result, a predetermined number of similarity score vectors in each face recognition algorithm as similarity values
- Step S306 performing fusion processing on the initial recognition result of the at least one face recognition algorithm according to the similarity value and the weight value, to obtain a recognition result of identifying the preset image.
- sorting results are obtained by sorting the similarity score vectors of each face recognition algorithm in at least one face recognition algorithm, wherein the sorting results may be performed in descending order.
- the number of similarity score vectors selected by each face recognition algorithm in at least one face recognition algorithm is identical, by multiplying each similarity score vector Si by a corresponding weight value Wi, And summation processing according to the identity letter, obtaining a similarity score vector of multiple identities, and dividing by the sum UM_W of the weight of the ownership, the final recognition result of identifying the preset image is obtained.
- FIG. 4 is a flow chart of steps of an optional image recognition method according to an embodiment of the present application, as shown in FIG. 4, according to weight values of at least one face recognition algorithm.
- the similarity value is obtained by fusing the initial recognition result of the at least one face recognition algorithm, and obtaining the recognition result of identifying the preset image, comprising the following steps:
- Step S402 determining a similarity threshold of each face recognition algorithm in at least one face recognition algorithm
- Step S404 comparing the similarity score vector and the similarity threshold of each face recognition algorithm
- Step S406 if the similarity score vector is greater than or equal to the similarity threshold, determining the similarity score vector as the similarity value
- Step S408 performing fusion processing on the initial recognition result of the at least one face recognition algorithm according to the similarity value and the weight value, to obtain a recognition result of identifying the preset image.
- the similarity threshold may be preset. Alternatively, a fixed similarity threshold may be set, and a similarity threshold corresponding to each algorithm may also be set.
- the ranking result is obtained by sorting the similarity score vector of each face recognition algorithm in at least one face recognition algorithm, wherein the ranking can be performed in descending order And if the similarity score vector is greater than or equal to the similarity threshold, determining the similarity score vector as the similarity value, in which case, the similarity score selected by each of the face recognition algorithms in at least one of the face recognition algorithms The number of vectors may be inconsistent.
- a similarity score vector of multiple identities is obtained, and then divided by the weight of the ownership.
- SUM_W the final recognition result of the recognized preset image is obtained.
- the initial recognition result of the at least one face recognition algorithm is fused by the following formula to obtain a recognition result of the recognized preset image:
- the similarity value in the above calculation formula may be a similarity value obtained according to all the similarity score vectors, or may be a similarity value obtained by selecting any similarity score vector.
- the embodiment of the present application further provides an image recognition apparatus. It should be noted that the method steps provided in the foregoing Embodiment 1 can be performed in the image recognition apparatus provided in the embodiment of the present application.
- FIG. 5 is a schematic structural diagram of an image recognition apparatus according to an embodiment of the present application. As shown in FIG. 5, the apparatus includes the following modules: an acquisition module 50, an assignment module 52, a determination module 54, and an identification module 56, where
- the obtaining module 50 is configured to acquire at least one face recognition algorithm, and the assignment module 52 is configured to assign a weight to the at least one face recognition algorithm to determine a weight value of the at least one face recognition algorithm; the determining module 54 And determining, by comparing at least one face recognition algorithm, an initial recognition result of the preset image and a preset recognition result, determining a similarity value of the at least one face recognition algorithm; and the recognition module 56 is configured to be according to at least one type of person The weight value and the similarity value of the face recognition algorithm are used to fuse the initial recognition result of at least one face recognition algorithm to obtain a recognition result of the recognized preset image.
- the obtaining module 50, the assigning module 52, the determining module 54, and the identifying module 56 correspond to steps S102 to S108 in Embodiment 1, and the foregoing modules are the same as the examples and application scenarios implemented by the corresponding steps, but not It is limited to the content disclosed in the above embodiment 1. It should be noted that the above modules may be implemented as part of a device in a computer system such as a set of computer executable instructions.
- each of the foregoing modules may be implemented by software or hardware.
- the foregoing may be implemented by, but not limited to, the foregoing modules are located in the same processor; or Module bits are in any combination in different processors.
- the embodiment of the present application further provides a storage medium, where the storage medium includes a stored program, wherein the program performs any of the optional or preferred image recognition methods described above.
- the storage medium provided by the embodiment of the present application is configured to store a program for performing at least one face recognition algorithm; assigning weights to at least one face recognition algorithm respectively, and determining weights of at least one face recognition algorithm And determining a similarity value of the at least one face recognition algorithm by comparing at least one face recognition algorithm to identify an initial recognition result of the preset image and the preset recognition result; and according to the weight value of the at least one face recognition algorithm
- the similarity value is used to fuse the initial recognition result of at least one face recognition algorithm to obtain a recognition result of identifying the preset image.
- the storage medium provided by the embodiment of the present application is configured to store a program that performs the following functions: collecting a predetermined number of face image data in a current scene to obtain an image test set.
- the storage medium provided by the embodiment of the present application is configured to store a program for performing the following functions: sorting a plurality of face recognition algorithms according to the size of the test accuracy of the plurality of face recognition algorithms, and selecting at least one person from the sorting result.
- the face recognition algorithm or, by comparing the test accuracy of the plurality of face recognition algorithms with the preset threshold, determining at least one face recognition algorithm.
- the storage medium configured to store a program for performing the following functions: at least one face recognition algorithm, each having a weight value of 1; according to the arrangement order of at least one face recognition algorithm, for at least one type of person
- the face recognition algorithm assigns weights; according to the test accuracy of at least one face recognition algorithm, at least one face recognition algorithm is weighted.
- the storage medium configured to store a program that performs the following functions: sorting the similarity score vector of each face recognition algorithm in at least one face recognition algorithm to obtain a sort result; and determining, according to the sort result, each The predetermined number of similarity score vectors in the face recognition algorithm are similarity values; according to the similarity value and the weight value, the initial recognition result of the at least one face recognition algorithm is merged to obtain the recognition result of the recognized preset image. .
- the storage medium provided by the embodiment of the present application is configured to store a program for performing a function of: determining a similarity threshold of each face recognition algorithm in at least one face recognition algorithm; comparing a similarity score vector of each face recognition algorithm with The magnitude of the similarity threshold; if the similarity score vector is greater than or equal to the similarity threshold, the similarity score vector is determined as the similarity value; and the initial recognition result of the at least one face recognition algorithm is performed according to the similarity value and the weight value
- the fusion processing obtains a recognition result that identifies the preset image.
- the embodiment of the present application further provides a processor, where the processor is configured to run a program, wherein the program executes any optional or preferred image recognition method described above.
- the processor provided in this embodiment is configured to run a program that performs the following functions: acquiring at least one face recognition algorithm; assigning corresponding weights to at least one face recognition algorithm, and determining weights of at least one face recognition algorithm And determining a similarity value of the at least one face recognition algorithm by comparing at least one face recognition algorithm to identify an initial recognition result of the preset image and the preset recognition result; and according to the weight value of the at least one face recognition algorithm The similarity value is used to fuse the initial recognition result of at least one face recognition algorithm to obtain a recognition result of identifying the preset image.
- the processor provided in this embodiment of the present application is configured to run a program that performs the following functions: collecting a predetermined number of face image data in a current scene to obtain an image test set.
- the processor provided in this embodiment of the present application is configured to run a program that performs the following functions: sorting multiple face recognition algorithms according to the test accuracy of multiple face recognition algorithms, and selecting at least one person from the sorting results.
- the face recognition algorithm or, by comparing the test accuracy of the plurality of face recognition algorithms with the preset threshold, determining at least one face recognition algorithm.
- the processor provided by the embodiment of the present application is configured to run a program that performs the following functions: at least one face recognition algorithm, each having a weight value of 1; according to the arrangement order of at least one face recognition algorithm, for at least one type of person
- the face recognition algorithm assigns weights; according to the test accuracy of at least one face recognition algorithm, at least one face recognition algorithm is weighted.
- the processor provided in this embodiment of the present application is configured to run a program that performs the following functions: sorting similarity score vectors of each face recognition algorithm in at least one face recognition algorithm to obtain a sort result; and determining, according to the sort result, each The predetermined number of similarity score vectors in the face recognition algorithm are similarity values; according to the similarity value and the weight value, the initial recognition result of the at least one face recognition algorithm is merged to obtain the recognition result of the recognized preset image. .
- the processor provided in this embodiment of the present application is configured to run a program that performs the following functions: determining a similarity threshold of each face recognition algorithm in at least one face recognition algorithm; comparing the similarity score vector of each face recognition algorithm with The magnitude of the similarity threshold; if the similarity score vector is greater than or equal to the similarity threshold, the similarity score vector is determined as the similarity value; and the initial recognition result of the at least one face recognition algorithm is performed according to the similarity value and the weight value
- the fusion processing obtains a recognition result that identifies the preset image.
- the disclosed technical contents may be implemented in other manners.
- the device embodiments described above are only schematic.
- the division of the unit may be a logical function division.
- there may be another division manner for example, multiple units or components may be combined or may be Integrate into another system, or some features can be ignored or not executed.
- the mutual coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interface, unit or module, and may be electrical or otherwise.
- the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed to multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of the embodiment.
- each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.
- the above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.
- the integrated unit if implemented in the form of a software functional unit and sold or used as a standalone product, may be stored in a computer readable storage medium.
- a computer readable storage medium A number of instructions are included to cause a computer device (which may be a personal computer, server or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present application.
- the foregoing storage medium includes: a U disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a removable hard disk, a magnetic disk, or an optical disk, and the like. .
- the image recognition method and apparatus provided by at least some embodiments of the present application have the following beneficial effects: at least one face recognition algorithm is obtained; at least one face recognition algorithm is assigned a corresponding weight, and at least one is determined.
- the weighting value and the similarity value of the recognition algorithm are used to fuse the initial recognition result of at least one face recognition algorithm, and the recognition result of the preset image is obtained, thereby achieving the purpose of improving the recognition accuracy of the universal recognition algorithm, thereby realizing
- the technical effect of applying the algorithm of multiple common scenes to the specific scene is solved, and the problem of low recognition accuracy is solved when the recognition algorithm is used to identify the face image.
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Collating Specific Patterns (AREA)
Abstract
本申请公开了一种图像识别方法及装置。其中,该方法包括:获取至少一种人脸识别算法;分别对至少一种人脸识别算法赋对应的权重,确定至少一种人脸识别算法的权重值;通过比较至少一种人脸识别算法识别预设图像的初始识别结果与预设识别结果,确定至少一种人脸识别算法的相似度值;根据至少一种人脸识别算法的权重值和相似度值,对至少一种人脸识别算法的初始识别结果进行融合处理,得到识别预设图像的识别结果。本申请解决了采用多种识别算法对人脸图像进行识别时,存在识别准确性较低的技术问题。
Description
本申请要求于2018年4月19日提交中国专利局、申请号为201810355523.6、申请名称“图像识别方法及装置”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
本申请涉及人脸识别技术领域,具体而言,涉及一种图像识别方法及装置。
人脸识别是基于人的脸部特征信息进行身份识别的一种生物识别技术,属于现阶段应用比较广泛的人工智能生物识别方法,也叫做人像识别,面部识别。随着人工智能技术和计算机识别技术的发展,人脸识别技术已经广泛应用于各个领域,而且随着技术的进一步成熟和社会认同度的提高,人脸识别技术将应用在更多的领域。
但是,在人脸识别的应用过程中,不同的应用场景需要应用不同的人脸识别算法。例如,在办公室场景,光照条件稳定,人脸的光照均匀正常,一般不会出现过暗或者过亮的场景,但人口密度多,人脸遮挡的情况会经常发生,并且由于同事之间的交谈,所以人脸角度多样化,该场景下的人脸识别算法会针对人脸遮挡和人脸角度做优化,使该算法对人脸遮挡和人脸角度的鲁棒性会更好;而在室外监控场景下做人脸识别时,室外人口密度相对较小,所以人脸遮挡情况少,摄像头相对较高,图片中人脸角度相对变化小,但是室外由于不同的天气情况,光照不稳定,且中午光照过亮,夜间光照过暗,所以该应用场景下的人脸识别算法会针对人脸光照和亮度优化,使该算法对人脸光照和亮度的鲁棒性算法会更好。
目前,人脸识别系统主要是由人脸识别算法优化实现,现有的已经被普遍运用的识别算法有两种,一种是应用单一算法,但单一算法具有一定的局限性,在应用于具体的应用场景时,会存在场景迁移导致准确度下降的问题,并且,如果将单一的具体场景的算法应用于另一具体场景中,会出现场景变化导致准确度下降的问题。
另一种是提供了一种多算法人脸识别融合系统,根据不同模态的摄像头得到同一张人脸的不同模态的图片,然后每种模态应用特定的算法进行识别,这种方案的弊端在于多模态的图片采集会更复杂,成本更高,并且不同模态的相似度尺度不同,在分 数融合阶段难度更高。
基于上述存在的技术问题,目前需要提供一种可以融合多种人脸识别算法进行人脸识别的解决方案。
发明内容
本申请至少部分实施例提供了一种图像识别方法及装置,以至少解决采用多种识别算法对人脸图像进行识别时,存在识别准确性较低的技术问题。
根据本申请其中一实施例的一个方面,提供了一种图像识别方法,包括:获取至少一种人脸识别算法;分别对至少一种人脸识别算法赋对应的权重,确定至少一种人脸识别算法的权重值;通过比较至少一种人脸识别算法识别预设图像的初始识别结果与预设识别结果,确定至少一种人脸识别算法的相似度值;根据至少一种人脸识别算法的权重值和相似度值,对至少一种人脸识别算法的初始识别结果进行融合处理,得到识别预设图像的识别结果。
可选地,获取至少一种人脸识别算法,包括:根据多种人脸识别算法对图像测试集进行测试,确定多种人脸识别算法的测试准确率,其中,图像测试集中包括:预设数量的正样本对和预设数量的负样本对,通过如下公式计算得到测试准确率:ACC=(TP+TN)÷(NUM_P+NUM_N),其中,ACC为测试准确率,TP为所有正样本对中识别为同一个人的对数,TN为所有负样本对中识别为不同一个人的对数,NUM_P为正样本对的总数,NUM_N为负样本对的总数;根据多种人脸识别算法的测试准确率的大小,确定至少一种人脸识别算法。
可选地,在获取至少一种人脸识别算法之前,方法还包括:采集当前场景下的预定数量的人脸图像数据,得到图像测试集。
可选地,根据多种人脸识别算法的测试准确率的大小,确定至少一种人脸识别算法,包括:根据多种人脸识别算法的测试准确率的大小,对多种人脸识别算法进行排序,从排序结果中选取至少一种人脸识别算法;或者,通过比较多种人脸识别算法的测试准确率与预设阈值,确定至少一种人脸识别算法。
可选地,通过以下方式至少之一,分别对至少一种人脸识别算法赋对应的权重:对至少一种人脸识别算法,均赋权重值为1;根据至少一种人脸识别算法的排列顺序,对至少一种人脸识别算法赋权重;根据至少一种人脸识别算法的测试准确率大小,对至少一种人脸识别算法赋权重。
可选地,根据至少一种人脸识别算法的权重值和相似度值,对至少一种人脸识别 算法的初始识别结果进行融合处理,得到识别预设图像的识别结果,包括:对至少一种人脸识别算法中每种人脸识别算法的相似度分数向量进行排序,得到排序结果;根据排序结果,确定每种人脸识别算法中预定数量的相似度分数向量为相似度值;根据相似度值和权重值,对至少一种人脸识别算法的初始识别结果进行融合处理,得到识别预设图像的识别结果。
可选地,根据至少一种人脸识别算法的权重值和相似度值,对至少一种人脸识别算法的初始识别结果进行融合处理,得到识别预设图像的识别结果,包括:确定至少一种人脸识别算法中每种人脸识别算法的相似度阈值;比较每种人脸识别算法的相似度分数向量与相似度阈值的大小;若相似度分数向量大于等于相似度阈值,则将相似度分数向量确定为相似度值;根据相似度值和权重值,对至少一种人脸识别算法的初始识别结果进行融合处理,得到识别预设图像的识别结果。
可选地,通过如下公式对至少一种人脸识别算法的初始识别结果进行融合处理,得到识别预设图像的识别结果:S=(Si×Wi)÷SUM_W=(S1×W1+S2×W2+…+Sn×Wn)÷SUM_W,其中,S为识别结果,Si=[Si1,Si2,Si3...SiP],Si为相似度值,i为第i种人脸识别算法,P为待测试人员的数量,n为上述人脸识别算法的数量,Wi为权重值,SUM_W为所有权重值的总和。
根据本申请其中一实施例的另一方面,还提供了一种图像识别装置,包括:获取模块,设置为获取至少一种人脸识别算法;赋值模块,设置为分别对至少一种人脸识别算法赋对应的权重,确定至少一种人脸识别算法的权重值;确定模块,设置为通过比较至少一种人脸识别算法识别预设图像的初始识别结果与预设识别结果,确定至少一种人脸识别算法的相似度值;识别模块,设置为根据至少一种人脸识别算法的权重值和相似度值,对至少一种人脸识别算法的初始识别结果进行融合处理,得到识别预设图像的识别结果。
根据本申请其中一实施例的另一方面,还提供了一种存储介质,存储介质包括存储的程序,其中,程序执行上述任意一项可选或优选的图像识别方法。
根据本申请其中一实施例的另一方面,还提供了一种处理器,处理器设置为运行程序,其中,程序运行时执行上述任意一项可选或优选的图像识别方法。
在本申请至少部分实施例中,通过获取至少一种人脸识别算法;分别对至少一种人脸识别算法赋对应的权重,确定至少一种人脸识别算法的权重值;通过比较至少一种人脸识别算法识别预设图像的初始识别结果与预设识别结果,确定至少一种人脸识别算法的相似度值;根据至少一种人脸识别算法的权重值和相似度值,对至少一种人 脸识别算法的初始识别结果进行融合处理,得到识别预设图像的识别结果,达到了提高通用识别算法的识别准确性的目的,从而实现了将多个通用场景的算法融合应用到特定场景的技术效果,进而解决了采用多种识别算法对人脸图像进行识别时,存在识别准确性较低的技术问题。
此处所说明的附图用来提供对本申请的进一步理解,构成本申请的一部分,本申请的示意性实施例及其说明用于解释本申请,并不构成对本申请的不当限定。在附图中:
图1是根据本申请实施例的一种图像识别方法的步骤流程图;
图2是根据本申请实施例的一种可选的图像识别方法的步骤流程图;
图3是根据本申请实施例的一种可选的图像识别方法的步骤流程图;
图4是根据本申请实施例的一种可选的图像识别方法的步骤流程图;以及
图5是根据本申请实施例的一种图像识别装置的结构示意图。
为了使本技术领域的人员更好地理解本申请方案,下面将结合本申请实施例中的附图,对本申请实施例中的技术方案进行清楚、完整地描述,显然,所描述的实施例仅仅是本申请一部分的实施例,而不是全部的实施例。基于本申请中的实施例,本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例,都应当属于本申请保护的范围。
需要说明的是,本申请的说明书和权利要求书及上述附图中的术语“第一”、“第二”等是用于区别类似的对象,而不必用于描述特定的顺序或先后次序。应该理解这样使用的数据在适当情况下可以互换,以便这里描述的本申请的实施例能够以除了在这里图示或描述的那些以外的顺序实施。此外,术语“包括”和“具有”以及他们的任何变形,意图在于覆盖不排他的包含,例如,包含了一系列步骤或单元的过程、方法、系统、产品或设备不必限于清楚地列出的那些步骤或单元,而是可包括没有清楚地列出的或对于这些过程、方法、产品或设备固有的其它步骤或单元。
实施例1
根据本申请实施例,提供了一种图像识别方法实施例,需要说明的是,在附图的 流程图示出的步骤可以在诸如一组计算机可执行指令的计算机系统中执行,并且,虽然在流程图中示出了逻辑顺序,但是在某些情况下,可以以不同于此处的顺序执行所示出或描述的步骤。
本申请可以实现将应用多个不同的人脸识别算法,应用于某一特定场景进行人脸识别,并且应用场景广泛,适应性强,通过几个通用场景的识别算法融合应用到某一特定场景中,并且可以提高通用识别算法的识别准确性。
图1是根据本申请实施例的一种图像识别方法的步骤流程图,如图1所示,该方法包括如下步骤:
步骤S102,获取至少一种人脸识别算法。
可选的,在上述步骤S102中,在单一人脸识别场景下,若存在多个人脸识别算法,则需要对多个人脸识别算法在当前场景下排序和筛选,以获取至少一种人脸识别算法。
在一种可选的实施例中,可以通过多个人脸识别算法对当前场景的图像测试集(例如,人脸图像测试集)进行测试,根据每种人脸识别算法在该图像测试集上的测试准确率,按照由高至低的顺序,选取至少一种人脸识别算法。
步骤S104,分别对至少一种人脸识别算法赋对应的权重,确定至少一种人脸识别算法的权重值。
可选的,在上述步骤S104中,可以通过不同的方式对筛选出的至少一种人脸识别算法中的每种人脸识别算法进行赋权重,若筛选出n种人脸识别算法,则对n种人脸识别算法所赋的权重值可以为[W1,W2,W3...Wn],其中,W为权重值,1、2、3……n为1-n种人脸识别算法。
在一种可选的实施例中,通过以下方式至少之一,分别对至少一种人脸识别算法赋对应的权重:
方式一:对至少一种人脸识别算法,均赋权重值为1;
可选的,可以对所选的n种人脸识别算法中的每种人脸算法均匀的赋权重值为1,即,W1=W2=...=Wn=1。
方式二:根据至少一种人脸识别算法的排列顺序,对至少一种人脸识别算法赋权重;
可选的,可以对所选的n种人脸识别算法按照排列顺序,单调递减的赋权重值,即[W1,W2,W3...Wn]=[n,n-1,n-1...1]。
方式三:根据至少一种人脸识别算法的测试准确率大小,对至少一种人脸识别算法赋权重。
可选的,可以根据所选的n种人脸识别算法的准确率的大小与所赋权重的大小,得到所选的n种人脸识别算法的准确率与所赋权重的比值,从而确定n种人脸识别算法的权重值,即W1:W2:W3...:Wn=ACC1:ACC2:ACC3:...:ACCn。
步骤S106,通过比较至少一种人脸识别算法识别预设图像的初始识别结果与预设识别结果,确定至少一种人脸识别算法的相似度值;
步骤S108,根据至少一种人脸识别算法的权重值和相似度值,对至少一种人脸识别算法的初始识别结果进行融合处理,得到识别预设图像的识别结果。
可选的,在上述步骤S106中,初始识别结果为至少一种人脸识别算法中的每种人脸识别算法识别预设图像的识别结果,预设识别结果为预先设置的正确的识别结果。
在一种可选的实施例中,可以预先建立一个人脸图像数据库,该人脸数据库可以用于存储多个人员信息(其中,人员信息可以包括人员的身份标识)与多张测试图片,可选的,可以预设该人脸数据库有P个人员信息,则采用所选的n种人脸识别算法中的每种人脸识别算法测试某一张测试照片,可以得到P个相似度值,其中,P个相似度值均可以为0-1之间的任意数值。
例如,第i种人脸识别算法识别该张测试照片的初始识别结果,可以为[(Si1,身份1),(Si2,身份2),(Si3,身份3)...(SiP,身份P)]。
可选的,在上述步骤S108中,可以根据上述P个相似度值和n种人脸识别算法的权重值,对至少一种人脸识别算法的初始识别结果进行融合处理,得到识别预设图像的识别结果。
在一种可选的实施例中,可以通过将最终的识别结果中的相似度向量进行排序,从而确定相似度向量最高的一个人员的身份,为该张测试图片的识别结果。
通过上述步骤,可以实现通过多种不同的人脸识别算法进行融合处理,得到识别结果,提高了多个识别算法在当前特定场景下的识别准确性。
在本申请实施例中,通过获取至少一种人脸识别算法;分别对至少一种人脸识别算法赋对应的权重,确定至少一种人脸识别算法的权重值;通过比较至少一种人脸识别算法识别预设图像的初始识别结果与预设识别结果,确定至少一种人脸识别算法的相似度值;根据至少一种人脸识别算法的权重值和相似度值,对至少一种人脸识别算法的初始识别结果进行融合处理,得到识别预设图像的识别结果,达到了提高通用识 别算法的识别准确性的目的,从而实现了将多个通用场景的算法融合应用到特定场景的技术效果,进而解决了采用多种识别算法对人脸图像进行识别时,存在识别准确性较低的技术问题。
在一种可选的实施例中,图2是根据本申请实施例的一种可选的图像识别方法的步骤流程图,如图2所示,获取至少一种人脸识别算法,包括如下步骤:
步骤S202,根据多种人脸识别算法对图像测试集进行测试,确定多种人脸识别算法的测试准确率,其中,图像测试集中包括:预设数量的正样本对和预设数量的负样本对,通过如下公式计算得到测试准确率:ACC=(TP+TN)÷(NUM_P+NUM_N),其中,ACC为测试准确率,TP为所有正样本对中识别为同一个人的对数,TN为所有负样本对中识别为不同一个人的对数,NUM_P为正样本对的总数,NUM_N为负样本对的总数;
步骤S204,根据多种人脸识别算法的测试准确率的大小,确定至少一种人脸识别算法。
在一种可选的实施例中,在获取至少一种人脸识别算法之前,上述方法还包括如下步骤:采集当前场景下的预定数量的人脸图像数据,得到图像测试集。
可选的,上述人脸图像数据可以通过拍摄照片的形式来采集得到,在获取至少一种人脸识别算法之前,可以采集当前应用场景的人脸图像数据,例如,当前应用场景下可以有100人,每人5张照片,共500张图片,进而可以构造人脸图像测试集,其中,该人脸图像测试集包括:正样本对和负样本对各1000对。
可选的,上述人脸图像测试集中的正样本对和负样本可以通过以下方式得到:
在一种可选的实施例中,同一个人有两张不同的人脸图片,每人5张照片,故每人可以形成10对正样本对,由于当前应用场景下共有100人,因此,将得到10×100=1000对正样本对,确定为该场景下人脸图像测试集的1000对正样本对。
在一种可选的实施例中,两个不同人的人脸图片,每人一张图片,每人的每张图片和其他人的一张照片构成一对负样本对。则每人每张照片可以构造99×5=495对负样本对,则每人构造495×5=2475对负样本对,因此,共有100×2475÷2=123750对负样本对,其中,除以2是由于每对负样本对被计算了2次,最后,从这123750对负样本对中,随机抽取1000对负样本对,确定为该场景下人脸图像测试集的1000对负样本对。
在一种可选的实施例中,根据多种人脸识别算法的测试准确率的大小,确定至少一种人脸识别算法,包括如下步骤:
根据多种人脸识别算法的测试准确率的大小,对多种人脸识别算法进行排序,从排序结果中选取至少一种人脸识别算法;
或者,通过比较多种人脸识别算法的测试准确率与预设阈值,确定至少一种人脸识别算法。
可选的,在上述可选的实施方案中,可以根据多种人脸识别算法的测试准确率的大小,直接选择排序前n个人脸识别算法。
在另一种可选的实施例中,可以预先给定一个测试准确度阈值,若测试准确率大于等于测试准确度阈值的多种人脸识别算法有K个,若K大于等于n,则选取前n个人脸识别算法为上述至少一种人脸识别算法,若K小于n,则选取K个人脸识别算法为上述至少一种人脸识别算法。
在一种可选的实施例中,图3是根据本申请实施例的一种可选的图像识别方法的步骤流程图,如图3所示,根据至少一种人脸识别算法的权重值和相似度值,对至少一种人脸识别算法的初始识别结果进行融合处理,得到识别预设图像的识别结果,包括如下步骤:
步骤S302,对至少一种人脸识别算法中每种人脸识别算法的相似度分数向量进行排序,得到排序结果;
步骤S304,根据排序结果,确定每种人脸识别算法中预定数量的相似度分数向量为相似度值;
步骤S306,根据相似度值和权重值,对至少一种人脸识别算法的初始识别结果进行融合处理,得到识别预设图像的识别结果。
在上述步骤S302至步骤S306中,通过对至少一种人脸识别算法中每种人脸识别算法的相似度分数向量进行排序,得到排序结果,其中,可以按照从大到小的顺序进行排序,在此情况下,至少一种人脸识别算法中的每种人脸识别算法所选取的相似度分数向量的个数一致,通过将每个相似度分数向量Si乘以对应的权重值Wi,并根据身份信进行求和处理,得到多个身份的相似度分数向量,再除以所有权重值的和SUM_W,得到识别预设图像的最终识别结果。
在一种可选的实施例中,图4是根据本申请实施例的一种可选的图像识别方法的步骤流程图,如图4所示,根据至少一种人脸识别算法的权重值和相似度值,对至少一种人脸识别算法的初始识别结果进行融合处理,得到识别预设图像的识别结果,包括如下步骤:
步骤S402,确定至少一种人脸识别算法中每种人脸识别算法的相似度阈值;
步骤S404,比较每种人脸识别算法的相似度分数向量与相似度阈值的大小;
步骤S406,若相似度分数向量大于等于相似度阈值,则将相似度分数向量确定为相似度值;
步骤S408,根据相似度值和权重值,对至少一种人脸识别算法的初始识别结果进行融合处理,得到识别预设图像的识别结果。
在上述步骤S402至步骤S408中,可以预先设定相似度阈值,可选的,可以设置固定的一个相似度阈值,还可以设置每个算法对应的相似度阈值。
在一种可选的实施例中,通过对至少一种人脸识别算法中每种人脸识别算法的相似度分数向量进行排序,得到排序结果,其中,可以按照从大到小的顺序进行排序,若相似度分数向量大于等于相似度阈值,则将相似度分数向量确定为相似度值,在此情况下,至少一种人脸识别算法中的每种人脸识别算法所选取的相似度分数向量的个数可能不一致,通过将每个相似度分数向量Si乘以对应的权重值Wi,并根据身份信进行求和处理,得到多个身份的相似度分数向量,再除以所有权重值的和SUM_W,得到识别预设图像的最终识别结果。
在一种可选的实施例中,通过如下公式对至少一种人脸识别算法的初始识别结果进行融合处理,得到识别预设图像的识别结果:
S=(Si×Wi)÷SUM_W=(S1×W1+S2×W2+…+Sn×Wn)÷SUM_W,其中,S为识别结果,Si=[Si1,Si2,Si3...SiP],Si为相似度值,i为第i种人脸识别算法,P为待测试人员的数量,n为人脸识别算法的数量,Wi为权重值,SUM_W为所有权重值的总和。
可选的,上述计算公式中的相似度值,可以为根据所有相似度分数向量得到相似度值,也可以为选取的任意个相似度分数向量得到的相似度值。
实施例2
本申请实施例还提供了一种图像识别装置,需要说明的是,上述实施例1所提供的方法步骤,可以但不限于在本申请实施例所提供的图像识别装置中执行。
图5是根据本申请实施例的一种图像识别装置的结构示意图,如图5所示,该装置包括以下模块:获取模块50、赋值模块52、确定模块54以及识别模块56,其中,
获取模块50,设置为获取至少一种人脸识别算法;赋值模块52,设置为分别对至 少一种人脸识别算法赋对应的权重,确定至少一种人脸识别算法的权重值;确定模块54,设置为通过比较至少一种人脸识别算法识别预设图像的初始识别结果与预设识别结果,确定至少一种人脸识别算法的相似度值;识别模块56,设置为根据至少一种人脸识别算法的权重值和相似度值,对至少一种人脸识别算法的初始识别结果进行融合处理,得到识别预设图像的识别结果。
需要说明的是,获取模块50、赋值模块52、确定模块54以及识别模块56对应于实施例1中的步骤S102至步骤S108,上述模块与对应的步骤所实现的示例和应用场景相同,但不限于上述实施例1所公开的内容。需要说明的是,上述模块作为装置的一部分可以在诸如一组计算机可执行指令的计算机系统中执行。
仍需要说明的是,上述各个模块是可以通过软件或硬件的形式来实现的,对于后者,可以通过以下方式来实现,但不限于此:上述各个模块位于同一处理器中;或者,上述各个模块位以任意组合的形式于不同的处理器中。
实施例3
本申请实施例还提供了一种存储介质,存储介质包括存储的程序,其中,程序执行上述任意一项可选或优选的图像识别方法。
本申请实施例提供的存储介质设置为存储执行以下功能的程序:获取至少一种人脸识别算法;分别对至少一种人脸识别算法赋对应的权重,确定至少一种人脸识别算法的权重值;通过比较至少一种人脸识别算法识别预设图像的初始识别结果与预设识别结果,确定至少一种人脸识别算法的相似度值;根据至少一种人脸识别算法的权重值和相似度值,对至少一种人脸识别算法的初始识别结果进行融合处理,得到识别预设图像的识别结果。
本申请实施例提供的存储介质设置为存储执行以下功能的程序:根据多种人脸识别算法对图像测试集进行测试,确定多种人脸识别算法的测试准确率,其中,图像测试集中包括:预设数量的正样本对和预设数量的负样本对,通过如下公式计算得到测试准确率:ACC=(TP+TN)÷(NUM_P+NUM_N),其中,ACC为测试准确率,TP为所有正样本对中识别为同一个人的对数,TN为所有负样本对中识别为不同一个人的对数,NUM_P为正样本对的总数,NUM_N为负样本对的总数;根据多种人脸识别算法的测试准确率的大小,确定至少一种人脸识别算法。
本申请实施例提供的存储介质设置为存储执行以下功能的程序:采集当前场景下的预定数量的人脸图像数据,得到图像测试集。
本申请实施例提供的存储介质设置为存储执行以下功能的程序:根据多种人脸识 别算法的测试准确率的大小,对多种人脸识别算法进行排序,从排序结果中选取至少一种人脸识别算法;或者,通过比较多种人脸识别算法的测试准确率与预设阈值,确定至少一种人脸识别算法。
本申请实施例提供的存储介质设置为存储执行以下功能的程序:对至少一种人脸识别算法,均赋权重值为1;根据至少一种人脸识别算法的排列顺序,对至少一种人脸识别算法赋权重;根据至少一种人脸识别算法的测试准确率大小,对至少一种人脸识别算法赋权重。
本申请实施例提供的存储介质设置为存储执行以下功能的程序:对至少一种人脸识别算法中每种人脸识别算法的相似度分数向量进行排序,得到排序结果;根据排序结果,确定每种人脸识别算法中预定数量的相似度分数向量为相似度值;根据相似度值和权重值,对至少一种人脸识别算法的初始识别结果进行融合处理,得到识别预设图像的识别结果。
本申请实施例提供的存储介质设置为存储执行以下功能的程序:确定至少一种人脸识别算法中每种人脸识别算法的相似度阈值;比较每种人脸识别算法的相似度分数向量与相似度阈值的大小;若相似度分数向量大于等于相似度阈值,则将相似度分数向量确定为相似度值;根据相似度值和权重值,对至少一种人脸识别算法的初始识别结果进行融合处理,得到识别预设图像的识别结果。
实施例4
本申请实施例还提供了一种处理器,处理器设置为运行程序,其中,程序运行时执行上述任意一项可选或优选的图像识别方法。
本申请实施例提供的处理器设置为运行执行以下功能的程序:获取至少一种人脸识别算法;分别对至少一种人脸识别算法赋对应的权重,确定至少一种人脸识别算法的权重值;通过比较至少一种人脸识别算法识别预设图像的初始识别结果与预设识别结果,确定至少一种人脸识别算法的相似度值;根据至少一种人脸识别算法的权重值和相似度值,对至少一种人脸识别算法的初始识别结果进行融合处理,得到识别预设图像的识别结果。
本申请实施例提供的处理器设置为运行执行以下功能的程序:根据多种人脸识别算法对图像测试集进行测试,确定多种人脸识别算法的测试准确率,其中,图像测试集中包括:预设数量的正样本对和预设数量的负样本对,通过如下公式计算得到测试准确率:ACC=(TP+TN)÷(NUM_P+NUM_N),其中,ACC为测试准确率,TP为所有正样本对中识别为同一个人的对数,TN为所有负样本对中识别为不同一个人的对数, NUM_P为正样本对的总数,NUM_N为负样本对的总数;根据多种人脸识别算法的测试准确率的大小,确定至少一种人脸识别算法。
本申请实施例提供的处理器设置为运行执行以下功能的程序:采集当前场景下的预定数量的人脸图像数据,得到图像测试集。
本申请实施例提供的处理器设置为运行执行以下功能的程序:根据多种人脸识别算法的测试准确率的大小,对多种人脸识别算法进行排序,从排序结果中选取至少一种人脸识别算法;或者,通过比较多种人脸识别算法的测试准确率与预设阈值,确定至少一种人脸识别算法。
本申请实施例提供的处理器设置为运行执行以下功能的程序:对至少一种人脸识别算法,均赋权重值为1;根据至少一种人脸识别算法的排列顺序,对至少一种人脸识别算法赋权重;根据至少一种人脸识别算法的测试准确率大小,对至少一种人脸识别算法赋权重。
本申请实施例提供的处理器设置为运行执行以下功能的程序:对至少一种人脸识别算法中每种人脸识别算法的相似度分数向量进行排序,得到排序结果;根据排序结果,确定每种人脸识别算法中预定数量的相似度分数向量为相似度值;根据相似度值和权重值,对至少一种人脸识别算法的初始识别结果进行融合处理,得到识别预设图像的识别结果。
本申请实施例提供的处理器设置为运行执行以下功能的程序:确定至少一种人脸识别算法中每种人脸识别算法的相似度阈值;比较每种人脸识别算法的相似度分数向量与相似度阈值的大小;若相似度分数向量大于等于相似度阈值,则将相似度分数向量确定为相似度值;根据相似度值和权重值,对至少一种人脸识别算法的初始识别结果进行融合处理,得到识别预设图像的识别结果。
上述本申请实施例序号仅仅为了描述,不代表实施例的优劣。
在本申请的上述实施例中,对各个实施例的描述都各有侧重,某个实施例中没有详述的部分,可以参见其他实施例的相关描述。
在本申请所提供的几个实施例中,应该理解到,所揭露的技术内容,可通过其它的方式实现。其中,以上所描述的装置实施例仅仅是示意性的,例如所述单元的划分,可以为一种逻辑功能划分,实际实现时可以有另外的划分方式,例如多个单元或组件可以结合或者可以集成到另一个系统,或一些特征可以忽略,或不执行。另一点,所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口,单元或模块的间接耦合或通信连接,可以是电性或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的,作为单元显示的部件可以是或者也可以不是物理单元,即可以位于一个地方,或者也可以分布到多个单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外,在本申请各个实施例中的各功能单元可以集成在一个处理单元中,也可以是各个单元单独物理存在,也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现,也可以采用软件功能单元的形式实现。
所述集成的单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时,可以存储在一个计算机可读取存储介质中。基于这样的理解,本申请的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的全部或部分可以以软件产品的形式体现出来,该计算机软件产品存储在一个存储介质中,包括若干指令用以使得一台计算机设备(可为个人计算机、服务器或者网络设备等)执行本申请各个实施例所述方法的全部或部分步骤。而前述的存储介质包括:U盘、只读存储器(ROM,Read-Only Memory)、随机存取存储器(RAM,Random Access Memory)、移动硬盘、磁碟或者光盘等各种可以存储程序代码的介质。
以上所述仅是本申请的优选实施方式,应当指出,对于本技术领域的普通技术人员来说,在不脱离本申请原理的前提下,还可以做出若干改进和润饰,这些改进和润饰也应视为本申请的保护范围。
如上所述,本申请至少部分实施例提供的图像识别方法及装置具有以下有益效果:通过获取至少一种人脸识别算法;分别对至少一种人脸识别算法赋对应的权重,确定至少一种人脸识别算法的权重值;通过比较至少一种人脸识别算法识别预设图像的初始识别结果与预设识别结果,确定至少一种人脸识别算法的相似度值;根据至少一种人脸识别算法的权重值和相似度值,对至少一种人脸识别算法的初始识别结果进行融合处理,得到识别预设图像的识别结果,达到了提高通用识别算法的识别准确性的目的,从而实现了将多个通用场景的算法融合应用到特定场景的技术效果,进而解决了采用多种识别算法对人脸图像进行识别时,存在识别准确性较低的技术问题。
Claims (14)
- 一种图像识别方法,包括:获取至少一种人脸识别算法;分别对所述至少一种人脸识别算法赋对应的权重,确定所述至少一种人脸识别算法的权重值;通过比较所述至少一种人脸识别算法识别预设图像的初始识别结果与预设识别结果,确定所述至少一种人脸识别算法的相似度值;根据所述至少一种人脸识别算法的权重值和相似度值,对所述至少一种人脸识别算法的初始识别结果进行融合处理,得到识别所述预设图像的识别结果。
- 根据权利要求1所述的方法,其中,获取至少一种人脸识别算法,包括:根据多种人脸识别算法对图像测试集进行测试,确定所述多种人脸识别算法的测试准确率,其中,所述图像测试集中包括:预设数量的正样本对和预设数量的负样本对,通过如下公式计算得到所述测试准确率:ACC=(TP+TN)÷(NUM_P+NUM_N),其中,ACC为所述测试准确率,TP为所有正样本对中识别为同一个人的对数,TN为所有负样本对中识别为不同一个人的对数,NUM_P为所述正样本对的总数,NUM_N为所述负样本对的总数;根据所述多种人脸识别算法的测试准确率的大小,确定所述至少一种人脸识别算法。
- 根据权利要求2所述的方法,其中,在获取至少一种人脸识别算法之前,所述方法还包括:采集当前场景下的预定数量的人脸图像数据,得到所述图像测试集。
- 根据权利要求2所述的方法,其中,根据所述多种人脸识别算法的测试准确率的大小,确定所述至少一种人脸识别算法,包括:根据所述多种人脸识别算法的测试准确率的大小,对所述多种人脸识别算法进行排序,从排序结果中选取所述至少一种人脸识别算法;或者,通过比较所述多种人脸识别算法的测试准确率与预设阈值,确定所述至少一 种人脸识别算法。
- 根据权利要求1所述的方法,其中,分别对所述至少一种人脸识别算法赋对应的权重:对所述至少一种人脸识别算法,均赋权重值为1。
- 根据权利要求1所述的方法,其中,分别对所述至少一种人脸识别算法赋对应的权重:根据所述至少一种人脸识别算法的排列顺序,对所述至少一种人脸识别算法赋权重。
- 根据权利要求1所述的方法,其中,分别对所述至少一种人脸识别算法赋对应的权重:根据所述至少一种人脸识别算法的测试准确率大小,对所述至少一种人脸识别算法赋权重。
- 根据权利要求5至7中任意一项所述的方法,其中,根据所述至少一种人脸识别算法的权重值和相似度值,对所述至少一种人脸识别算法的初始识别结果进行融合处理,得到识别所述预设图像的识别结果,包括:对所述至少一种人脸识别算法中每种人脸识别算法的相似度分数向量进行排序,得到排序结果;根据所述排序结果,确定所述每种人脸识别算法中预定数量的相似度分数向量为所述相似度值;根据所述相似度值和所述权重值,对所述至少一种人脸识别算法的初始识别结果进行融合处理,得到识别所述预设图像的识别结果。
- 根据权利要求5至7中任意一项所述的方法,其中,根据所述至少一种人脸识别算法的权重值和相似度值,对所述至少一种人脸识别算法的初始识别结果进行融合处理,得到识别所述预设图像的识别结果,包括:确定所述至少一种人脸识别算法中每种人脸识别算法的相似度阈值;比较所述每种人脸识别算法的相似度分数向量与相似度阈值的大小;若所述相似度分数向量大于等于所述相似度阈值,则将所述相似度分数向量确定为所述相似度值;根据所述相似度值和所述权重值,对所述至少一种人脸识别算法的初始识别结果进行融合处理,得到识别所述预设图像的识别结果。
- 根据权利要求8所述的方法,其中,通过如下公式对所述至少一种人脸识别算法的初始识别结果进行融合处理,得到识别所述预设图像的识别结果:S=(Si×Wi)÷SUM_W=(S1×W1+S2×W2+…+Sn×Wn)÷SUM_W,其中,S为所述识别结果,Si为所述相似度值,Si=[Si1,Si2,Si3...SiP],i为第i种人脸识别算法,P为待测试人员的数量,n为所述人脸识别算法的数量,Wi为所述权重值,SUM_W为所有权重值的总和。
- 一种图像识别装置,包括:获取模块,设置为获取至少一种人脸识别算法;赋值模块,设置为分别对所述至少一种人脸识别算法赋对应的权重,确定所述至少一种人脸识别算法的权重值;确定模块,设置为通过比较所述至少一种人脸识别算法识别预设图像的初始识别结果与预设识别结果,确定所述至少一种人脸识别算法的相似度值;识别模块,设置为根据所述至少一种人脸识别算法的权重值和相似度值,对所述至少一种人脸识别算法的初始识别结果进行融合处理,得到识别所述预设图像的识别结果。
- 一种存储介质,所述存储介质包括存储的程序,其中,所述程序执行权利要求1至10中任意一项所述的图像识别方法。
- 一种处理器,所述处理器设置为运行程序,其中,所述程序运行时执行权利要求1至10中任意一项所述的图像识别方法。
- 根据权利要求9所述的方法,其中,通过如下公式对所述至少一种人脸识别算法的初始识别结果进行融合处理,得到识别所述预设图像的识别结果:S=(Si×Wi)÷SUM_W=(S1×W1+S2×W2+…+Sn×Wn)÷SUM_W,其中,S为所述识别结果,Si为所述相似度值,Si=[Si1,Si2,Si3...SiP],i为第i种人脸识别算法,P为待测试人员的数量,n为所述人脸识别算法的数量,Wi为所述权重值,SUM_W为所有权重值的总和。
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810355523.6 | 2018-04-19 | ||
| CN201810355523.6A CN108776768A (zh) | 2018-04-19 | 2018-04-19 | 图像识别方法及装置 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2019200902A1 true WO2019200902A1 (zh) | 2019-10-24 |
Family
ID=64026458
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2018/115612 Ceased WO2019200902A1 (zh) | 2018-04-19 | 2018-11-15 | 图像识别方法及装置 |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN108776768A (zh) |
| WO (1) | WO2019200902A1 (zh) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110942065A (zh) * | 2019-11-26 | 2020-03-31 | Oppo广东移动通信有限公司 | 文本框选方法、装置、终端设备及计算机可读存储介质 |
| CN111242047A (zh) * | 2020-01-15 | 2020-06-05 | 北京明略软件系统有限公司 | 图像处理方法和装置、电子设备及计算机可读存储介质 |
| CN111444849A (zh) * | 2020-03-27 | 2020-07-24 | 上海依图网络科技有限公司 | 人员识别方法、装置、电子设备、及计算机可读存储介质 |
| CN111538503A (zh) * | 2020-03-27 | 2020-08-14 | 北京达佳互联信息技术有限公司 | 安装包体积的优化方法、装置、电子设备及存储介质 |
| CN112182265A (zh) * | 2020-09-17 | 2021-01-05 | 国家电网有限公司 | 广谱集成多算法的图像识别方法 |
| CN112597886A (zh) * | 2020-12-22 | 2021-04-02 | 成都商汤科技有限公司 | 乘车逃票检测方法及装置、电子设备和存储介质 |
| CN112597823A (zh) * | 2020-12-07 | 2021-04-02 | 深延科技(北京)有限公司 | 注意力识别方法、装置、电子设备及存储介质 |
| CN114049659A (zh) * | 2021-09-24 | 2022-02-15 | 公安部第一研究所 | 一种1比n人脸识别引擎多算法融合方法 |
| CN114328657A (zh) * | 2021-12-21 | 2022-04-12 | 中国农业银行股份有限公司 | 一种去匿名化方法、装置、电子设备及存储介质 |
| CN114821252A (zh) * | 2022-03-16 | 2022-07-29 | 电子科技大学 | 一种图像识别算法的自成长方法 |
| CN116434313A (zh) * | 2023-04-28 | 2023-07-14 | 北京声迅电子股份有限公司 | 基于多人脸识别模块的人脸识别方法 |
| CN116721452A (zh) * | 2023-06-13 | 2023-09-08 | 光控特斯联(上海)信息科技有限公司 | 一种人脸识别系统、方法、装置、电子设备及程序产品 |
Families Citing this family (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108776768A (zh) * | 2018-04-19 | 2018-11-09 | 广州视源电子科技股份有限公司 | 图像识别方法及装置 |
| CN111400817B (zh) * | 2018-12-30 | 2023-04-04 | 长城汽车股份有限公司 | 模块化结构的汽车车架连接点确定方法与非承载式汽车车架和汽车 |
| CN111400821B (zh) * | 2018-12-30 | 2023-04-07 | 长城汽车股份有限公司 | 长度或宽度可调的汽车车架连接点确定方法与非承载式汽车车架和汽车 |
| CN109800699A (zh) * | 2019-01-15 | 2019-05-24 | 珠海格力电器股份有限公司 | 图像识别方法、系统及装置 |
| CN109871762B (zh) * | 2019-01-16 | 2023-08-08 | 平安科技(深圳)有限公司 | 一种人脸识别模型的评价方法和装置 |
| CN110008815A (zh) * | 2019-01-25 | 2019-07-12 | 平安科技(深圳)有限公司 | 人脸识别融合模型的生成方法和装置 |
| CN110334695A (zh) * | 2019-07-18 | 2019-10-15 | 国网电子商务有限公司 | 一种基于人脸特征的身份认证方法及装置 |
| CN111178259A (zh) * | 2019-12-30 | 2020-05-19 | 八维通科技有限公司 | 支持多算法融合的识别方法及系统 |
| CN111126340B (zh) * | 2019-12-31 | 2023-12-01 | 北京旷视科技有限公司 | 对象识别方法、装置、电子设备及计算机可读存储介质 |
| CN111325250B (zh) * | 2019-12-31 | 2023-08-01 | 深圳云天励飞技术有限公司 | 图像处理方法及相关设备 |
| CN112712026B (zh) * | 2020-12-29 | 2023-08-01 | 广州广电运通智能科技有限公司 | 身份识别方法、人脸注册方法、计算机设备和存储介质 |
| CN113095672B (zh) * | 2021-04-09 | 2024-06-07 | 公安部物证鉴定中心 | 一种人脸图像比对算法评测方法和系统 |
| CN119579540B (zh) * | 2024-11-16 | 2025-11-14 | 合肥工业大学 | 基于弱监督多模态对比学习的预测模型构建方法及乳腺癌her2评分预测方法 |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106250858A (zh) * | 2016-08-05 | 2016-12-21 | 重庆中科云丛科技有限公司 | 一种融合多种人脸识别算法的识别方法及系统 |
| CN106650693A (zh) * | 2016-12-30 | 2017-05-10 | 河北三川科技有限公司 | 一种用于人脸比对的多特征融合识别算法 |
| CN107563280A (zh) * | 2017-07-24 | 2018-01-09 | 南京道熵信息技术有限公司 | 基于多模型的人脸识别方法和装置 |
| CN108776768A (zh) * | 2018-04-19 | 2018-11-09 | 广州视源电子科技股份有限公司 | 图像识别方法及装置 |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7773800B2 (en) * | 2001-06-06 | 2010-08-10 | Ying Liu | Attrasoft image retrieval |
| KR100543707B1 (ko) * | 2003-12-04 | 2006-01-20 | 삼성전자주식회사 | 서브그룹별 pca 학습을 이용한 얼굴인식방법 및 장치 |
| CN102262729B (zh) * | 2011-08-03 | 2013-01-02 | 山东志华信息科技股份有限公司 | 基于集成学习的混合融合人脸识别方法 |
| CN104123560B (zh) * | 2014-07-03 | 2017-12-01 | 中山大学 | 基于相位编码特征和多度量学习的模糊人脸图像验证方法 |
| CN107633242A (zh) * | 2017-10-23 | 2018-01-26 | 广州视源电子科技股份有限公司 | 网络模型的训练方法、装置、设备和存储介质 |
-
2018
- 2018-04-19 CN CN201810355523.6A patent/CN108776768A/zh active Pending
- 2018-11-15 WO PCT/CN2018/115612 patent/WO2019200902A1/zh not_active Ceased
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106250858A (zh) * | 2016-08-05 | 2016-12-21 | 重庆中科云丛科技有限公司 | 一种融合多种人脸识别算法的识别方法及系统 |
| CN106650693A (zh) * | 2016-12-30 | 2017-05-10 | 河北三川科技有限公司 | 一种用于人脸比对的多特征融合识别算法 |
| CN107563280A (zh) * | 2017-07-24 | 2018-01-09 | 南京道熵信息技术有限公司 | 基于多模型的人脸识别方法和装置 |
| CN108776768A (zh) * | 2018-04-19 | 2018-11-09 | 广州视源电子科技股份有限公司 | 图像识别方法及装置 |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110942065B (zh) * | 2019-11-26 | 2023-12-12 | Oppo广东移动通信有限公司 | 文本框选方法、装置、终端设备及计算机可读存储介质 |
| CN110942065A (zh) * | 2019-11-26 | 2020-03-31 | Oppo广东移动通信有限公司 | 文本框选方法、装置、终端设备及计算机可读存储介质 |
| CN111242047A (zh) * | 2020-01-15 | 2020-06-05 | 北京明略软件系统有限公司 | 图像处理方法和装置、电子设备及计算机可读存储介质 |
| CN111444849A (zh) * | 2020-03-27 | 2020-07-24 | 上海依图网络科技有限公司 | 人员识别方法、装置、电子设备、及计算机可读存储介质 |
| CN111538503A (zh) * | 2020-03-27 | 2020-08-14 | 北京达佳互联信息技术有限公司 | 安装包体积的优化方法、装置、电子设备及存储介质 |
| CN111444849B (zh) * | 2020-03-27 | 2024-02-27 | 上海依图网络科技有限公司 | 人员识别方法、装置、电子设备、及计算机可读存储介质 |
| CN112182265A (zh) * | 2020-09-17 | 2021-01-05 | 国家电网有限公司 | 广谱集成多算法的图像识别方法 |
| CN112597823A (zh) * | 2020-12-07 | 2021-04-02 | 深延科技(北京)有限公司 | 注意力识别方法、装置、电子设备及存储介质 |
| CN112597886A (zh) * | 2020-12-22 | 2021-04-02 | 成都商汤科技有限公司 | 乘车逃票检测方法及装置、电子设备和存储介质 |
| CN114049659A (zh) * | 2021-09-24 | 2022-02-15 | 公安部第一研究所 | 一种1比n人脸识别引擎多算法融合方法 |
| CN114049659B (zh) * | 2021-09-24 | 2024-04-19 | 公安部第一研究所 | 一种1比n人脸识别引擎多算法融合方法 |
| CN114328657A (zh) * | 2021-12-21 | 2022-04-12 | 中国农业银行股份有限公司 | 一种去匿名化方法、装置、电子设备及存储介质 |
| CN114821252B (zh) * | 2022-03-16 | 2023-05-26 | 电子科技大学 | 一种图像识别算法的自成长方法 |
| CN114821252A (zh) * | 2022-03-16 | 2022-07-29 | 电子科技大学 | 一种图像识别算法的自成长方法 |
| CN116434313B (zh) * | 2023-04-28 | 2023-11-14 | 北京声迅电子股份有限公司 | 基于多人脸识别模块的人脸识别方法 |
| CN116434313A (zh) * | 2023-04-28 | 2023-07-14 | 北京声迅电子股份有限公司 | 基于多人脸识别模块的人脸识别方法 |
| CN116721452A (zh) * | 2023-06-13 | 2023-09-08 | 光控特斯联(上海)信息科技有限公司 | 一种人脸识别系统、方法、装置、电子设备及程序产品 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108776768A (zh) | 2018-11-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2019200902A1 (zh) | 图像识别方法及装置 | |
| US9247106B2 (en) | Color correction based on multiple images | |
| CN103617432B (zh) | 一种场景识别方法及装置 | |
| CN110941594B (zh) | 一种视频文件的拆分方法、装置、电子设备及存储介质 | |
| US11281939B2 (en) | Method and apparatus for training an object identification neural network, and computer device | |
| CN112084903A (zh) | 一种更新人脸识别底库照片的方法及系统 | |
| CN110472494A (zh) | 脸部特征提取模型训练方法、脸部特征提取方法、装置、设备及存储介质 | |
| CN107609493A (zh) | 优化人脸图片质量评价模型的方法及装置 | |
| US8842889B1 (en) | System and method for automatic face recognition | |
| CN109284733A (zh) | 一种基于yolo和多任务卷积神经网络的导购消极行为监控方法 | |
| CN112508094A (zh) | 垃圾图片的识别方法、装置及设备 | |
| CN112306829B (zh) | 性能信息的确定方法及装置、存储介质、终端 | |
| CN113792827B (zh) | 目标对象识别方法、电子设备和计算机可读存储介质 | |
| CN111445442B (zh) | 基于神经网络的人群计数方法、装置、服务器及存储介质 | |
| CN109472193A (zh) | 人脸检测方法及装置 | |
| CN117011563B (zh) | 基于半监督联邦学习的道路损害巡检跨域检测方法及系统 | |
| CN112084904A (zh) | 人脸搜索方法、装置及存储介质 | |
| TW201604821A (zh) | 協作性人臉標註方法以及協作性人臉標註系統 | |
| CN111814617B (zh) | 基于视频的火灾判定方法、装置、计算机设备及存储介质 | |
| CN109711328A (zh) | 一种人脸识别方法、装置及电子设备 | |
| JP3962517B2 (ja) | 顔面検出方法及びその装置、コンピュータ可読媒体 | |
| WO2022193232A1 (zh) | 人脸聚类方法及装置、分类存储方法、介质、电子设备 | |
| CN114549502B (zh) | 一种人脸质量评估的方法、装置、电子设备及存储介质 | |
| CN116310630A (zh) | 调色模型训练方法、处理方法及其装置、设备及存储介质 | |
| WO2025248456A1 (en) | Computer-implemented method for jointly training at least two different spoof-detection systems |
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: 18915514 Country of ref document: EP Kind code of ref document: A1 |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 18915514 Country of ref document: EP Kind code of ref document: A1 |