WO2019037346A1 - 优化人脸图片质量评价模型的方法及装置 - Google Patents

优化人脸图片质量评价模型的方法及装置 Download PDF

Info

Publication number
WO2019037346A1
WO2019037346A1 PCT/CN2017/116217 CN2017116217W WO2019037346A1 WO 2019037346 A1 WO2019037346 A1 WO 2019037346A1 CN 2017116217 W CN2017116217 W CN 2017116217W WO 2019037346 A1 WO2019037346 A1 WO 2019037346A1
Authority
WO
WIPO (PCT)
Prior art keywords
face image
face
tested
similarity
identity information
Prior art date
Application number
PCT/CN2017/116217
Other languages
English (en)
French (fr)
Inventor
陈�全
Original Assignee
广州视源电子科技股份有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 广州视源电子科技股份有限公司 filed Critical 广州视源电子科技股份有限公司
Publication of WO2019037346A1 publication Critical patent/WO2019037346A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis

Definitions

  • the present invention relates to the field of image analysis technology, and in particular, to a method, device, storage medium and computer device for optimizing a face picture quality evaluation model.
  • the accuracy of the existing face image quality evaluation method in the actual non-limiting application scenario is much lower than the accuracy in the experiment.
  • the reason is that the general face image quality evaluation method is subjective selection of face images. Different features or attributes, such as the number of pixels in a unit image, to evaluate the quality of the picture.
  • the recognition accuracy of the face image that is subjective to the human eye in the face recognition algorithm is not necessarily high. Therefore, the existing face picture quality evaluation model has a poor evaluation effect on the face picture quality.
  • the present invention provides a method and a device for optimizing a face picture quality evaluation model, which can improve the objectivity of a face picture quality evaluation model for image quality evaluation, and thereby better evaluate the quality of a face picture.
  • the inventive solution includes:
  • an embodiment of the present invention provides a method for optimizing a face image quality evaluation model, including:
  • the test set includes a plurality of face images to be tested, and each face image to be tested is marked with corresponding first identity information;
  • the face database includes a plurality of sample face images, and each sample face image is marked with corresponding second identity information;
  • the regression neural network is used to train the initial face image quality evaluation model and parameters, and the optimized face image quality evaluation model and parameters are obtained.
  • an embodiment of the present invention provides an apparatus for optimizing a face image quality evaluation model, including:
  • test set establishing module configured to set a face image test set;
  • the test set includes a plurality of face images to be tested, and each face image to be tested is marked with corresponding first identity information;
  • An initial recognition module configured to identify a similarity between the face image to be tested and the sample face image in the preset face database, and obtain each face to be tested according to the similarity, the first identity information, and the second identity information a recognition result of the image;
  • the face database includes a plurality of sample face images, and each sample face image is marked with corresponding second identity information;
  • a quality score calculation module configured to determine, according to the recognition result, a quality score of each face image to be tested
  • the training module is used to train the initial face image quality evaluation model and parameters by using the regression model and the corresponding quality score as training data, and obtain an optimized face image quality evaluation model and parameter.
  • a computer readable storage medium having stored thereon a computer program that, when executed by a processor, implements the steps of the method.
  • a computer apparatus comprising a memory, a processor, and a computer program stored on the memory and operative on the processor, the processor implementing the steps of the method when the program is executed.
  • the test set includes a plurality of face images to be tested, inputting a pre-established face database; and identifying a similarity between the face image to be tested and the sample face image in the preset face database, according to
  • the similarity and the identity information of each picture are obtained by the recognition result of each face image to be tested, and the recognition result of each face image to be tested is obtained; the quality score of each face image to be tested is determined to test the concentrated
  • the face image and its corresponding quality score are used for neural network training of training data.
  • the face image quality evaluation model and parameters obtained through training are more objectively evaluated by the face image quality evaluation model and parameters. Accurate, overcoming the influence of human subjective factors on the evaluation of picture quality.
  • FIG. 1 is a schematic flowchart of a method for optimizing a face picture quality evaluation model according to an embodiment
  • FIG. 2 is a schematic flowchart of a method for optimizing a face picture quality evaluation model in a specific application scenario
  • FIG. 3 is a schematic structural diagram of an apparatus for optimizing a face picture quality evaluation model according to an embodiment.
  • FIG. 1 is a schematic flow chart of a method for optimizing a face picture quality evaluation model according to an embodiment; As shown in FIG. 1, the method for optimizing a face picture quality evaluation model in this embodiment includes the steps of:
  • a face image test set is established, where the test set includes a plurality of face images to be tested, and each face image to be tested is set with corresponding first identity information.
  • the face image to be tested in the face image test set is a face image collected in a specific application scenario, and can cover multiple people. Optionally, multiple pictures per person are taken from different angles.
  • the identity information of the face image refers to information that can uniquely identify the identity of the person, such as the name of the person and the identification number of the person.
  • the preset face database includes face images of a plurality of individuals, and each picture is marked with corresponding second identity information. Optionally, one picture per person.
  • Comparing the face image to be tested with the sample face image in the preset face database is to calculate the similarity between the two faces, and the face image and the sample to be tested can be determined according to the similarity and the preset similarity threshold. Whether the face image is the same person.
  • the similarity between the face image to be tested and the sample face image in the preset face database is identified, and each face to be tested is obtained according to the similarity, the first identity information, and the second identity information.
  • the specific manner of the recognition result of the picture includes: respectively calculating the similarity between the face image to be tested and the face image of each sample in the face database, and obtaining the maximum similarity value and the maximum similarity of each face image to be tested.
  • Corresponding sample face image; obtaining first identity information of the face image to be tested, and acquiring second identity information of the corresponding sample face image; according to the maximum value of the similarity, the preset similarity threshold, the first The identity information and the second identity information obtain the recognition result of each face image to be tested.
  • the recognition result includes a type of result, a second type of result, three types of results, and four types of results;
  • a type of result is that the face image to be tested and the sample face image having the same identity information are correctly identified as the same face;
  • the second type of result is the wrong face image and sample to be tested with the same identity information.
  • the face image is recognized as a different face;
  • the three types of results are wrong to identify the face image and the sample face image with different identity information as the same face;
  • the four types of results are correct and will have different identity information.
  • the face image and the sample face image are identified as different faces.
  • the face recognition algorithm is preset to calculate the similarity between the face image to be tested and the face image of each sample in the face database, and obtain the maximum similarity of each face image to be tested. The value, and the corresponding sample face image when the similarity is maximum.
  • a sample face image with the greatest similarity to the face image to be tested can be obtained (also referred to as a comparison sample face). Picture) and the corresponding similarity (ie the maximum similarity). And further identifying whether the face image to be tested and the face image of the comparison sample are the same face.
  • the recognition result of each face image to be tested is obtained according to the maximum value of the similarity, the preset similarity threshold, and the corresponding identity information.
  • the model and network parameters of the regression neural network are obtained when the preset convergence condition is met, thereby obtaining an optimized face image quality evaluation model and parameters.
  • the method for optimizing the face image quality evaluation model of the above embodiment first establishes a face a picture test set; then, according to the face picture test set and the preset face database, the recognition result of each face image to be tested is obtained; and then, the quality score of each face image to be tested is determined according to the recognition result;
  • the neural network training is carried out with the test face image and its corresponding face quality score in the test set as the training data, and the optimized face image quality evaluation model and parameters are obtained.
  • the face image quality evaluation model and parameters can be optimized, and the optimized face image quality evaluation model and parameters are adopted.
  • the evaluation of face image quality will not be affected by subjective factors, and the evaluation result is more objective and accurate.
  • the face recognition algorithm separately calculates the similarity between the face image to be tested and the face image of each sample in the face database, including: separately extracting face features and samples of each face image to be tested.
  • the face feature of the face image is calculated based on the extracted face features.
  • the recognition result of each face image to be tested is obtained according to the maximum value of the similarity, the preset similarity threshold, and the corresponding identity information, including: the maximum value of the similarity and the pre- Comparing the similarity thresholds to obtain a first comparison result; comparing the first identity information with the second identity information to obtain a second comparison result; according to the first comparison result, the second comparison As a result, the recognition result of each face image to be tested is obtained.
  • the identification result includes: a type of result, a second type of result, a third type of result, and four types of results; wherein, in the one type of result and the three types of results, the maximum value of the similarity is greater than or equal to the similarity threshold; In the second type of results and the four types of results, the maximum similarity is smaller than the similarity threshold.
  • the implementation manner of obtaining the identification result is specifically as follows:
  • test_name If score ⁇ threshold, and name_labeled is different from test_name, it is identified as four types of results;
  • the score indicates the maximum degree of similarity corresponding to the face image to be tested
  • the threshold indicates the similarity threshold
  • the name_labeled indicates the identity information corresponding to the sample face image
  • the test_name indicates the identity information corresponding to the face image to be tested.
  • the manner in which the quality score is calculated may be:
  • FIG. 2 is a schematic flow chart of a method for optimizing a face picture quality evaluation model according to a specific embodiment, which is specifically divided into four parts: data set finishing S21, face recognition S22, face quality labeling S23, and face quality.
  • the method for optimizing a face image quality evaluation model in this embodiment includes the following steps:
  • the data collating part includes: the setting of the face recognition algorithm face_identify_algorithm, the preset of the face database face_gallery_set, and the establishment of the face test set face_test_set.
  • the similarity threshold is used to determine whether two face images belong to one person; and the preset face database includes multiple sample face images (for example, N sheets) , N is greater than 1), each sample face picture is set with corresponding second identity information; Optional, there is only one picture per person.
  • the established face test set includes a plurality of face images to be tested, and each face image to be tested is set with corresponding first identity information; optionally, the face image to be tested is collected in a face image of a plurality of different angles. . It can be understood that the first identity information set by the face image to be tested and the second identity information set by the sample face image are the same type of information, for example, both are names.
  • the face recognition portion determines the recognition result of the face image to be tested according to the similarity threshold threshold.
  • the specific method is: extracting feature information of each test face image (assuming that the name of the current test face image is marked as test_name) by using a face recognition algorithm (the feature information includes both subjective perception features) Information, such as pixels, brightness, etc.; also includes other feature information that is not perceived by humans); the feature information of the extracted test face image is compared with the feature information corresponding to N sample face images in the face database, respectively, N similarities are obtained; the N similarity results are sorted, and the similarity score score with the largest similarity and the name_labeled of the corresponding sample face image are taken, and the following recognition result identification_result is obtained:
  • FP (ie, the second type of result): if score ⁇ threshold, but name_labeled is the same as test_name;
  • TN (ie four types of results): if score ⁇ threshold, and name_labeled is different from test_name.
  • One type of result is the correct face image and sample face image that will have the same identity information. Recognized as the same face; the second type of result is wrong to identify the face image and the sample face image with the same identity information as different faces; the three types of results are the wrong faces with different identity information.
  • the picture and the sample face picture are recognized as the same face; the four types of results are correct to identify the face image and the sample face picture with different identity information as different faces.
  • the quality score of the face test picture is obtained according to the recognition result of step S22.
  • the face quality annotation is to add the calculated face quality score to the face image test set to obtain a face quality test set, which is a test picture set containing the calculated face quality score.
  • This process uses regression neural network training data to obtain regression neural network parameters.
  • the regression neural network model can be regarded as the face image quality evaluation model to be optimized, the parameters in the regression neural network and the parameters of the face image quality evaluation model, and the neural network training is to train the face image quality.
  • the evaluation model and parameters that is, the regression neural network is used to train the parameters involved in the face image quality evaluation model.
  • the preset convergence condition is met, the regression neural network model and parameters are obtained, thereby obtaining optimized Face image quality evaluation model and parameters.
  • face_identify_algorithm face_gallery_set, face_test_set, and threshold are used in the above technical features to represent a face recognition algorithm, a face database, a face picture test set, and a similarity threshold.
  • the above representation should not be construed as a limitation on the face recognition algorithm, the face database, the face picture test set, and the similarity threshold.
  • the present invention also provides an apparatus for optimizing a face picture quality evaluation model, which can be used to perform the above method for optimizing a face picture quality evaluation model.
  • an apparatus for optimizing a face picture quality evaluation model which can be used to perform the above method for optimizing a face picture quality evaluation model.
  • the illustrated structure does not constitute a limitation to the device, and may include more or fewer components than those illustrated, or a combination of certain components, or a different component arrangement.
  • FIG. 3 is a schematic structural diagram of an apparatus for optimizing a face picture quality evaluation model according to an embodiment of the present invention.
  • the apparatus for optimizing a face picture quality evaluation model in this embodiment includes: a test set establishing module 310.
  • the initial identification module 320, the quality score calculation module 330, and the training module 340 are detailed as follows:
  • the test set establishing module 310 is configured to establish a face image test set; the test set includes a plurality of face images to be tested, and each face image to be tested is set with corresponding first identity information;
  • the initial identification module 320 is configured to identify a similarity between the face image to be tested and the sample face image in the preset face database, and obtain the respective similarities according to the similarity, the first identity information, and the second identity information. Determining a recognition result of the face image; the face database includes a plurality of sample face images, and each sample face image is marked with corresponding second identity information;
  • the quality score calculation module 330 is configured to determine a quality score of each face image to be tested according to the recognition result
  • the training module 340 is configured to train the initial face image quality evaluation model and parameters by using a regression neural network with the image of the face to be tested and the corresponding quality score as the training data, and obtain the optimized face image quality. Evaluate models and parameters.
  • the initial recognition module 320 is configured to separately calculate the similarity between the face image to be tested and the face image of each sample in the face database. Obtaining a maximum similarity value corresponding to each face image to be tested, and a corresponding sample face image when the similarity maximum value is obtained; acquiring first identity information of the face image to be tested, and acquiring a second corresponding face image of the sample face Identity information: obtaining, according to the maximum value of the similarity, the preset similarity threshold, the first identity information, and the second identity information, a recognition result of each face image to be tested.
  • the initial recognition module 320 may include a similarity calculation unit and a face recognition unit.
  • the similarity calculation unit is configured to separately calculate the similarity between the face image to be tested and the face image of each sample in the face database, and obtain the maximum similarity and similarity of each face image to be tested. The corresponding sample face image at the maximum value.
  • the face recognition unit is configured to obtain a recognition result of each face image to be tested according to the maximum value of the similarity, the preset similarity threshold, and the identity information corresponding to the picture.
  • the similarity calculation unit may separately extract a face feature of each face image to be tested, a face feature of the sample face image, and calculate a similarity based on the extracted face feature.
  • the face recognition unit is configured to compare the maximum value of the similarity with a preset similarity threshold to obtain a first comparison result; and the first identity information and The second identity information is compared to obtain a second comparison result; and the recognition result of each face image to be tested is obtained according to the first comparison result and the second comparison result.
  • the recognition result of each face image to be tested is obtained as follows:
  • test_name If score ⁇ threshold, and name_labeled is different from test_name, it is identified as four types of results;
  • score represents the maximum similarity of the face image to be tested
  • threshold represents phase
  • name_labeled represents the identity information corresponding to the sample face image
  • test_name represents the identity information corresponding to the face image to be tested.
  • the quality score calculation module 330 is configured to extract a similarity maximum value and a similarity threshold corresponding to each recognition result; and determine, according to the recognition result, the similarity maximum value, and the similarity threshold, The quality score of each face image to be tested.
  • the quality score calculation module 330 can calculate the quality score of each face image to be tested by the following formula:
  • each face information corresponding to the identity information is at least two; in the face database, The sample face image corresponding to each identity information is one.
  • each program module is only an example, and the actual application may be implemented according to requirements, for example, according to configuration requirements of the corresponding hardware or software.
  • the above function assignment is performed by different program modules, that is, the internal structure of the device for optimizing the face picture quality evaluation model is divided into different processes.
  • the module is programmed to perform all or part of the functions described above.
  • the storage medium may also be disposed in a computer device, where the computer device further includes a processor, and when the processor executes the program in the storage medium, all of the embodiments of the foregoing methods can be implemented. Or part of the steps.
  • the storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Image Analysis (AREA)
  • Collating Specific Patterns (AREA)

Abstract

一种优化人脸图片质量评价模型的方法及装置,所述方法包括:建立人脸图片测试集;识别待测人脸图片与预设人脸数据库中的样本人脸图片的相似度,根据所述相似度以及图片身份信息得出各待测人脸图片的识别结果;根据所述识别结果确定各待测人脸图片的质量分数;以待测人脸图片及其对应的质量分数为训练数据进行神经网络训练,得到优化的人脸图片质量评价模型及参数。通过上述人脸图片质量评价模型及参数,对人脸图片质量的评价将不受人为主观因素的影响。

Description

优化人脸图片质量评价模型的方法及装置 技术领域
本发明涉及图像分析技术领域,特别是涉及优化人脸图片质量评价模型的方法、装置、存储介质及计算机设备。
背景技术
随着深度学习和人脸识别技术的发展,人脸识别已经应用于越来越多的场景去快速的识别一个人的身份。现有的人脸图片质量评价方法在实际非限制性的应用场景中的准确率会比实验中的准确率低很多,究其原因:一般的人脸图片质量评价方法是主观的挑选人脸图片的不同特征或者属性,如,单位图像中像素的数量来评价图片的质量。但是在实际应用场景中,符合人眼主观好的人脸图片在人脸识别算法中的识别准确率并不一定高。因此,现有人脸图片质量评价模型对人脸图片质量的评价效果较差。
发明内容
基于此,本发明提供了优化人脸图片质量评价模型的方法及装置,能够提高人脸图片质量评价模型对图片质量评价的客观性,进而更好地评价出人脸图片的质量。
本发明方案包括:
第一方面,本发明实施例提供一种优化人脸图片质量评价模型的方法,包括:
建立人脸图片测试集;测试集中包括多个待测人脸图片,各待测人脸图片标注有对应的第一身份信息;
识别待测人脸图片与预设人脸数据库中的样本人脸图片的相似度,根据所述相似度、第一身份信息和第二身份信息得出各待测人脸图片的识别结果;所 述人脸数据库中包括多个样本人脸图片,各样本人脸图片标注有对应的第二身份信息;
根据所述识别结果确定各待测人脸图片的质量分数;
以各待测人脸图片及其对应的质量分数为训练数据,采用回归型神经网络对初始人脸图片质量评价模型及参数进行训练,得到优化的人脸图片质量评价模型及参数。
第二方面,本发明实施例提供一种优化人脸图片质量评价模型的装置,包括:
测试集建立模块,用于建立人脸图片测试集;测试集中包括多个待测人脸图片,各待测人脸图片标注有对应的第一身份信息;
初始识别模块,用于识别待测人脸图片与预设人脸数据库中的样本人脸图片的相似度,根据所述相似度、第一身份信息和第二身份信息得出各待测人脸图片的识别结果;所述人脸数据库中包括多个样本人脸图片,各样本人脸图片标注有对应的第二身份信息;
质量分数计算模块,用于根据所述识别结果确定出各待测人脸图片的质量分数;以及,
训练模块,用于以各待测人脸图片及其对应的质量分数为训练数据,采用回归型神经网络对初始人脸图片质量评价模型及参数进行训练,得到优化的人脸图片质量评价模型及参数。
一种计算机可读存储介质,其上存储有计算机程序,该程序被处理器执行时实现所述方法的步骤。
一种计算机设备,包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序,所述处理器执行所述程序时实现所述方法的步骤。
本发明实施例提供的技术方案带来的有益效果:
通过建立人脸图片测试集,测试集中包括多个待测人脸图片,输入预先建立的人脸数据库;识别待测人脸图片与预设人脸数据库中的样本人脸图片的相似度,根据所述相似度以及各图片的身份信息得出各待测人脸图片的识别结果得出各待测人脸图片的识别结果;确定出各待测人脸图片的质量分数,以测试集中的待测人脸图片及其对应的质量分数为训练数据进行神经网络训练,通过训练得到的人脸图片质量评价模型及参数,通过该人脸图片质量评价模型及参数对人脸图片质量的评价更客观准确,克服了人为主观因素对图片质量的评价影响。
附图说明
图1为一实施例的优化人脸图片质量评价模型的方法的示意性流程图;
图2为一具体应用场景下的优化人脸图片质量评价模型的方法的示意性流程图;
图3为一实施例的优化人脸图片质量评价模型的装置的示意性结构图。
具体实施方式
为了使本发明的目的、技术方案及优点更加清楚明白,以下结合附图及实施例,对本发明进行进一步详细说明。应当理解,此处所描述的具体实施例仅仅用以解释本发明,并不用于限定本发明。
本发明中的步骤虽然用标号进行了排列,但并不用于限定步骤的先后次序,除非明确说明了步骤的次序或者某步骤的执行需要其他步骤作为基础,否则步骤的相对次序是可以调整的。
图1为一实施例的优化人脸图片质量评价模型的方法的示意性流程图;如 图1所示,本实施例中的优化人脸图片质量评价模型的方法包括步骤:
S11,建立人脸图片测试集,所述测试集中包括多个待测人脸图片,各待测人脸图片设置有对应的第一身份信息。
人脸图片测试集中的待测人脸图片是在具体应用场景下采集的人脸图片,可以覆盖多个人。可选地,每人多张图片,取自不同的角度。人脸图片的身份信息指的是可以唯一识别人身份的信息,例如可以是人物的姓名、身边识别号码。
S12,识别待测人脸图片与预设人脸数据库中的样本人脸图片的相似度,根据所述相似度、第一身份信息和第二身份信息得出各待测人脸图片的识别结果。
其中,预设人脸数据库包含若干个人的人脸图片,并且每张图片标注有对应的第二身份信息。可选地,每人一张图片。
将待测人脸图片与预设人脸数据库中的样本人脸图片进行比对,是计算两个人脸的相似度,根据相似度和预设的相似度阈值可判断待测人脸图片和样本人脸图片是否为同一个人。
在一实施例中,识别待测人脸图片与预设人脸数据库中的样本人脸图片的相似度,根据所述相似度、第一身份信息和第二身份信息得出各待测人脸图片的识别结果的具体方式包括:分别计算待测人脸图片与人脸数据库中各样本人脸图片的相似度,得出各待测人脸图片对应的相似度最大值,以及相似度最大值时对应的样本人脸图片;获取待测人脸图片的第一身份信息,获取对应的样本人脸图片的第二身份信息;根据所述相似度最大值、预设的相似度阈值、第一身份信息和第二身份信息得到各待测人脸图片的识别结果。
可选地,所述识别结果包括一类结果、二类结果、三类结果和四类结果; 具体例如:一类结果为正确的将具有相同身份信息的待测人脸图片和样本人脸图片识别为同一人脸;二类结果为错误的将具有相同身份信息的待测人脸图片和样本人脸图片识别为不同人脸;三类结果为错误的将具有不同身份信息的待测人脸图片和样本人脸图片识别为同一人脸;四类结果为正确的将具有不同身份信息的待测人脸图片和样本人脸图片识别为不同人脸。
在一可选实施例中,预设人脸识别算法,分别计算待测人脸图片与人脸数据库中各样本人脸图片的相似度,并得出各待测人脸图片对应的相似度最大值,以及相似度最大值时对应的样本人脸图片。通过计算各待测人脸图片与人脸数据库中每一个样本人脸图片的相似度,可以得到与所述待测人脸图片相似度最大的一个样本人脸图片(也称为比较样本人脸图片)及对应的相似度(即相似度最大值)。进而识别所述待测人脸图片与所述比较样本人脸图片是否为同一人脸。可选的,根据所述相似度最大值、预设的相似度阈值以及对应的身份信息,得到各待测人脸图片的识别结果。
S13,根据所述识别结果确定各待测人脸图片的质量分数。
可选的,根据所述识别结果、相似度最大值和相似度阈值,确定各待测人脸图片的质量分数,以便于进行神经网络训练。
S14,以测试集中的待测人脸图片及其对应的质量分数为训练数据,采用回归型神经网络对人脸图片质量评价模型及参数进行训练,得到优化的人脸图片质量评价模型及参数。
可以理解的,采用回归型神经网络进行训练时,当满足预设收敛条件时获取所述回归型神经网络的模型及网络参数,由此可得到优化的人脸图片质量评价模型及参数。
综上,上述实施例的优化人脸图片质量评价模型的方法,首先,建立人脸 图片测试集;然后,根据人脸图片测试集和预设人脸数据库得出各待测人脸图片的识别结果;继而,根据所述识别结果确定各待测人脸图片的质量分数;最后,以测试集中的待测人脸图片及其对应的人脸质量分数为训练数据进行神经网络训练,得到优化的人脸图片质量评价模型及参数。由此能够优化人脸图片质量评价模型及参数,采用优化后的人脸图片质量评价模型及参数,对人脸图片质量的评价将不受人为主观因素的影响,评价结果更客观更准确。
在一可选实施例中,通过人脸识别算法分别计算待测人脸图片与人脸数据库中各样本人脸图片的相似度,包括:分别提取各待测人脸图片的人脸特征、样本人脸图片的人脸特征,基于提取的人脸特征计算相似度。
在一可选实施例中,根据所述相似度最大值、预设的相似度阈值以及对应的身份信息,得到各待测人脸图片的识别结果,包括:将所述相似度最大值和预设的相似度阈值进行比对,得到第一比对结果;将所述第一身份信息和第二身份信息进行比对,得到第二比对结果;根据第一比对结果、第二比对结果得到各待测人脸图片的识别结果。可选地,所述识别结果包括:一类结果、二类结果、三类结果和四类结果;其中,一类结果和三类结果中,相似度最大值大于或等于所述相似度阈值;二类结果和四类结果中,相似度最大值小于所述相似度阈值。
可选地,上述得到识别结果的实现方式具体例如:
如果score>=threshold,且name_labeled与test_name相同,则识别为一类结果;
如果score<threshold,但name_labeled与test_name相同,则识别为二类结果;
如果score>=threshold,但name_labeled与test_name不同,则识别为三 类结果;
如果score<threshold,且name_labeled与test_name不同,则识别为四类结果;
其中,score表示待测人脸图片所对应的相似度最大值,threshold表示相似度阈值,name_labeled表示样本人脸图片对应的身份信息,test_name表示待测人脸图片对应的身份信息。
在一可选实施例中,计算质量分数的方式可为:
quality_score=total+total×flag×|score–threshold|/delta;
Figure PCTCN2017116217-appb-000001
其中,2*total表示质量分数的满分值;如果score<threshold,则delta=threshold,否则delta=1-threshold。例如,公式中的total为50时,则质量分数的满分值为100。
图2为一具体实施例下的优化人脸图片质量评价模型的方法的示意性流程图,具体分为四个部分:数据集整理S21、人脸识别S22、人脸质量标注S23和人脸质量算法学习S24。
请参考如图2,本实施例中的优化人脸图片质量评价模型的方法包括步骤:
S21、数据整理。
数据整理部分包括:人脸识别算法face_identify_algorithm的设置、人脸数据库face_gallery_set的预设及人脸测试集face_test_set的建立。
对人脸识别算法设置对应的相似度阈值threshold,该相似度阈值用于判断两张人脸图片是否属于一个人;所述预设的人脸数据库中包括多个样本人脸图片(例如N张,N大于1),每张样本人脸图片设置有对应的第二身份信息; 可选的,每人只有一张图片。建立的人脸测试集中包括多个待测人脸图片,每张待测人脸图片设置有对应的第一身份信息;可选的,待测人脸图片采集于多人不同角度的人脸图片。可以理解的,待测人脸图片设置的第一身份信息与样本人脸图片设置的第二身份信息为同一类型的信息,例如,均为姓名。样本人脸图片的姓名name_labeled,待测人脸图片的姓名test_name。
S22、人脸识别。
人脸识别部分为根据相似度阈值threshold,确定待测人脸图片的识别结果。
在一实施例中,具体做法为:用人脸识别算法提取每张测试人脸图片(假设当前这张测试人脸图片的姓名标注为test_name)的特征信息(所述特征信息既包括主观感知的特征信息,如像素、亮度等;也包括其他不被人为感知的特征信息);使用提取的测试人脸图片的特征信息分别与人脸数据库中的N张样本人脸图片对应的特征信息进行对比,得到N个相似度;将这N个相似度结果排序,取出相似度最大的相似度分数score和对应的样本人脸图片的姓名name_labeled,进而可得到如下识别结果identify_result:
TP(即一类结果):如果score>=threshold,并且name_labeled与test_name相同;
FP(即二类结果):如果score<threshold,但是name_labeled与test_name相同;
FN(即三类结果):如果score>=threshold,但是name_labeled与test_name不同;
TN(即四类结果):如果score<threshold,并且name_labeled与test_name不同。
一类结果为正确的将具有相同身份信息的待测人脸图片和样本人脸图片 识别为同一人脸;二类结果为错误的将具有相同身份信息的待测人脸图片和样本人脸图片识别为不同人脸;三类结果为错误的将具有不同身份信息的待测人脸图片和样本人脸图片识别为同一人脸;四类结果为正确的将具有不同身份信息的待测人脸图片和样本人脸图片识别为不同人脸。
S23、人脸质量标注。
根据步骤S22的识别结果得到人脸测试图片的质量分数。人脸质量标注是将计算出来的人脸质量分数添加到人脸图片测试集中,得到人脸质量测试集,所述人脸质量测试集为包含计算出来的人脸质量分数的测试图片集。
S24、人脸质量算法学习。
该过程使用回归型神经网络训练数据,得到回归型神经网络参数。具体来说,回归型神经网络模型可视为待优化的人脸图片质量评价模型,回归型神经网络中的参数及所述人脸图片质量评价模型的参数,神经网络训练即训练人脸图片质量评价模型及参数,即采用回归型神经网络对人脸图片质量评价模型中涉及的参数进行训练,当满足预设收敛条件时,获取所述回归型神经网络模型及参数,由此可得到优化的人脸图片质量评价模型及参数。
为了简单表示,上述技术特征中分别用face_identify_algorithm、face_gallery_set、face_test_set、threshold表示人脸识别算法、人脸数据库、人脸图片测试集和相似度阈值。上述表示方式不应当理解为对人脸识别算法、人脸数据库、人脸图片测试集和相似度阈值的限制。
基于与上述实施例中的优化人脸图片质量评价模型的方法相同的思想,本发明还提供优化人脸图片质量评价模型的装置,该装置可用于执行上述优化人脸图片质量评价模型的方法。为了便于说明,优化人脸图片质量评价模型的装置实施例的结构示意图中,仅仅示出了与本发明实施例相关的部分,本领域技 术人员可以理解,图示结构并不构成对装置的限定,可以包括比图示更多或更少的部件,或者组合某些部件,或者不同的部件布置。
图3为本发明一实施例的优化人脸图片质量评价模型的装置的示意性结构图,如图3所示,本实施例的优化人脸图片质量评价模型的装置包括:测试集建立模块310、初始识别模块320、质量分数计算模块330以及训练模块340,各模块详述如下:
所述测试集建立模块310,用于建立人脸图片测试集;测试集中包括多个待测人脸图片,各待测人脸图片设置有对应的第一身份信息;
所述初始识别模块320,用于识别待测人脸图片与预设人脸数据库中的样本人脸图片的相似度,根据所述相似度、第一身份信息和第二身份信息得出各待测人脸图片的识别结果;所述人脸数据库中包括多个样本人脸图片,各样本人脸图片标注有对应的第二身份信息;
所述质量分数计算模块330,用于根据所述识别结果确定出各待测人脸图片的质量分数;
所述训练模块340,用于以各待测人脸图片及其对应的质量分数为训练数据,采用回归型神经网络对初始人脸图片质量评价模型及参数进行训练,得到优化的人脸图片质量评价模型及参数。
在一可选实施例中,在优化人脸图片质量评价模型的装置中,所述初始识别模块320,用于分别计算待测人脸图片与人脸数据库中各样本人脸图片的相似度,得出各待测人脸图片对应的相似度最大值,以及相似度最大值时对应的样本人脸图片;获取待测人脸图片的第一身份信息,获取对应的样本人脸图片的第二身份信息;根据所述相似度最大值、预设的相似度阈值、第一身份信息和第二身份信息得到各待测人脸图片的识别结果。
在一可选实施例中,在所述优化人脸图片质量评价模型的装置中,所述初始识别模块320可包括相似度计算单元和人脸识别单元。
其中,所述相似度计算单元,用于分别计算待测人脸图片与人脸数据库中各样本人脸图片的相似度,得出各待测人脸图片对应的相似度最大值,以及相似度最大值时对应的样本人脸图片。
其中,所述人脸识别单元,用于根据所述相似度最大值、预设的相似度阈值以及图片对应的身份信息,得到各待测人脸图片的识别结果。
在一可选实施例中,所述相似度计算单元,可分别提取各待测人脸图片的人脸特征、样本人脸图片的人脸特征,基于提取的人脸特征计算相似度。
在一可选实施例中,所述人脸识别单元,用于将所述相似度最大值和预设的相似度阈值进行比对,得到第一比对结果;将所述第一身份信息和第二身份信息进行比对,得到第二比对结果;根据第一比对结果、第二比对结果得到各待测人脸图片的识别结果。具体可按照如下方式得到各待测人脸图片的识别结果,包括:
如果score>=threshold,且name_labeled与test_name相同,则识别为一类结果;
如果score<threshold,但name_labeled与test_name相同,则识别为二类结果;
如果score>=threshold,但name_labeled与test_name不同,则识别为三类结果;
如果score<threshold,且name_labeled与test_name不同,则识别为四类结果;
其中,score表示待测人脸图片所对应的相似度最大值,threshold表示相 似度阈值,name_labeled表示样本人脸图片对应的身份信息,test_name表示待测人脸图片对应的身份信息。
在一可选实施例中,所述质量分数计算模块330,用于提取各识别结果所对应的相似度最大值和相似度阈值;根据所述识别结果、相似度最大值和相似度阈值,确定各待测人脸图片的质量分数。
例如:所述质量分数计算模块330可通过如下公式计算各待测人脸图片的质量分数:
quality_score=total+total×flag×|score–threshold|/delta;
Figure PCTCN2017116217-appb-000002
其中,2*total表示质量分数的满分值;如果score<threshold,则delta=threshold,否则delta=1-threshold。
在一可选实施例中,上述优化人脸图片质量评价模型的装置实施例中,所述测试集中,每一身份信息对应的待测人脸图片为至少两张;所述人脸数据库中,每一身份信息对应的样本人脸图片为一张。
需要说明的是,上述示例的优化人脸图片质量评价模型的装置的实施方式中,各模块/单元之间的信息交互、执行过程等内容,由于与本发明前述方法实施例基于同一构思,其带来的技术效果与本发明前述方法实施例相同,具体内容可参见本发明方法实施例中的叙述,此处不再赘述。
此外,上述示例的优化人脸图片质量评价模型的装置的实施方式中,各程序模块的逻辑划分仅是举例说明,实际应用中可以根据需要,例如出于相应硬件的配置要求或者软件的实现的便利考虑,将上述功能分配由不同的程序模块完成,即将所述优化人脸图片质量评价模型的装置的内部结构划分成不同的程 序模块,以完成以上描述的全部或者部分功能。
本领域普通技术人员可以理解,实现上述实施例方法中的全部或部分流程,是可以通过计算机程序来指令相关的硬件来完成,所述的程序可存储于一计算机可读取存储介质中,作为独立的产品销售或使用。所述程序在执行时,可执行如上述各方法的实施例的全部或部分步骤。此外,所述存储介质还可设置于一种计算机设备中,所述计算机设备中还包括处理器,所述处理器执行所述存储介质中的程序时,能够实现上述各方法的实施例的全部或部分步骤。其中,所述的存储介质可为磁碟、光盘、只读存储记忆体(Read-Only Memory,ROM)或随机存储记忆体(Random Access Memory,RAM)等。
在上述实施例中,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,可以参见其它实施例的相关描述。可以理解,其中所使用的术语“第一”、“第二”等在本文中用于区分对象,但这些对象不受这些术语限制。
以上所述实施例仅表达了本发明的几种实施方式,不能理解为对本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,在不脱离本发明构思的前提下,还可以做出若干变形和改进,这些都属于本发明的保护范围。因此,本发明专利的保护范围应以所附权利要求为准。

Claims (10)

  1. 一种优化人脸图片质量评价模型的方法,其特征在于,包括:
    建立人脸图片测试集;测试集中包括多个待测人脸图片,各待测人脸图片标注有对应的第一身份信息;
    识别待测人脸图片与预设人脸数据库中的样本人脸图片的相似度,根据所述相似度、第一身份信息和第二身份信息得出各待测人脸图片的识别结果;所述人脸数据库中包括多个样本人脸图片,各样本人脸图片标注有对应的第二身份信息;
    根据所述识别结果确定各待测人脸图片的质量分数;
    以各待测人脸图片及其对应的质量分数为训练数据,采用回归型神经网络对初始人脸图片质量评价模型及参数进行训练,得到优化的人脸图片质量评价模型及参数。
  2. 根据权利要求1所述的优化人脸图片质量评价模型的方法,其特征在于,识别待测人脸图片与预设人脸数据库中的样本人脸图片的相似度,根据所述相似度、第一身份信息和第二身份信息得出各待测人脸图片的识别结果,包括:
    分别计算待测人脸图片与人脸数据库中各样本人脸图片的相似度,得出各待测人脸图片对应的相似度最大值,以及相似度最大值时对应的样本人脸图片;
    获取待测人脸图片的第一身份信息,获取对应的样本人脸图片的第二身份信息;根据所述相似度最大值、预设的相似度阈值、第一身份信息和第二身份信息,得到各待测人脸图片的识别结果。
  3. 根据权利要求2所述的优化人脸图片质量评价模型的方法,其特征在于,分别计算待测人脸图片与人脸数据库中各样本人脸图片的相似度,包括:
    分别提取各待测人脸图片的人脸特征、样本人脸图片的人脸特征,基于提取的人脸特征计算相似度。
  4. 根据权利要求2或3所述的优化人脸图片质量评价模型的方法,其特征在于,根据所述相似度最大值、预设的相似度阈值、第一身份信息和第二身份信息,得到各待测人脸图片的识别结果,包括:
    将所述相似度最大值和预设的相似度阈值进行比对,得到第一比对结果;
    将所述第一身份信息和第二身份信息进行比对,得到第二比对结果;
    根据第一比对结果、第二比对结果得到各待测人脸图片的识别结果。
  5. 根据权利要求4所述的优化人脸图片质量评价模型的方法,其特征在于,所述识别结果包括:一类结果、二类结果、三类结果和四类结果;其中,一类结果和三类结果中,相似度最大值大于或等于所述相似度阈值;二类结果和四类结果中,相似度最大值小于所述相似度阈值;
    根据所述识别结果确定各待测人脸图片的质量分数,包括:
    提取各识别结果所对应的相似度最大值和相似度阈值;
    根据所述识别结果、相似度最大值和相似度阈值,确定各待测人脸图片的质量分数。
  6. 根据权利要求5所述的优化人脸图片质量评价模型的方法,其特征在于,通过如下公式计算各待测人脸图片的质量分数:
    quality_score=total+total×flag×|score–threshold|/delta;
    Figure PCTCN2017116217-appb-100001
    其中,2*total表示质量分数的满分值;如果score<threshold,则delta=threshold,否则delta=1-threshold;quality_score表示质量分数,score 表示待测人脸图片所对应的相似度最大值,threshold表示相似度阈值。
  7. 根据权利要求1、2、3、5或6所述的优化人脸图片质量评价模型的方法,其特征在于,
    所述测试集中,同一身份信息对应的待测人脸图片为至少两张;所述人脸数据库中,同一身份信息对应的样本人脸图片为一张。
  8. 一种优化人脸图片质量评价模型的装置,其特征在于,包括:
    测试集建立模块,用于建立人脸图片测试集;测试集中包括多个待测人脸图片,各待测人脸图片标注有对应的第一身份信息;
    初始识别模块,用于识别待测人脸图片与预设人脸数据库中的样本人脸图片的相似度,根据所述相似度、第一身份信息和第二身份信息得出各待测人脸图片的识别结果;所述人脸数据库中包括多个样本人脸图片,各样本人脸图片标注有对应的第二身份信息;
    质量分数计算模块,用于根据所述识别结果确定出各待测人脸图片的质量分数;以及,
    训练模块,用于以各待测人脸图片及其对应的质量分数为训练数据,采用回归型神经网络对初始人脸图片质量评价模型及参数进行训练,得到优化的人脸图片质量评价模型及参数。
  9. 一种计算机可读存储介质,其上存储有计算机程序,其特征在于,该程序被处理器执行时实现权利要求1至7任一所述方法的步骤。
  10. 一种计算机设备,包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序,其特征在于,所述处理器执行所述程序时实现权利要求1至7任一所述方法的步骤。
PCT/CN2017/116217 2017-08-25 2017-12-14 优化人脸图片质量评价模型的方法及装置 WO2019037346A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710743257.X 2017-08-25
CN201710743257.XA CN107609493B (zh) 2017-08-25 2017-08-25 优化人脸图片质量评价模型的方法及装置

Publications (1)

Publication Number Publication Date
WO2019037346A1 true WO2019037346A1 (zh) 2019-02-28

Family

ID=61055766

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/116217 WO2019037346A1 (zh) 2017-08-25 2017-12-14 优化人脸图片质量评价模型的方法及装置

Country Status (2)

Country Link
CN (1) CN107609493B (zh)
WO (1) WO2019037346A1 (zh)

Cited By (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109977815A (zh) * 2019-03-13 2019-07-05 上海商汤智能科技有限公司 图像质量评价方法及装置、电子设备、存储介质
CN110866437A (zh) * 2019-09-23 2020-03-06 平安科技(深圳)有限公司 颜值判定模型优化方法、装置、电子设备及存储介质
CN111191584A (zh) * 2019-12-30 2020-05-22 电信科学技术第十研究所有限公司 一种人脸识别方法及装置
CN111311581A (zh) * 2020-02-20 2020-06-19 杭州涂鸦信息技术有限公司 基于光照的图片评分方法及其系统和装置
CN111382681A (zh) * 2020-02-28 2020-07-07 浙江大华技术股份有限公司 一种人脸注册方法、装置及存储介质
CN111382693A (zh) * 2020-03-05 2020-07-07 北京迈格威科技有限公司 图像质量的确定方法、装置、电子设备和计算机可读介质
CN111797845A (zh) * 2019-03-19 2020-10-20 北京沃东天骏信息技术有限公司 图片处理方法、装置、存储介质及电子设备
CN111814570A (zh) * 2020-06-12 2020-10-23 深圳禾思众成科技有限公司 一种基于动态阈值的人脸识别方法、系统及存储介质
CN111966852A (zh) * 2020-06-28 2020-11-20 北京百度网讯科技有限公司 基于人脸的虚拟整容的方法和装置
CN112148908A (zh) * 2020-10-23 2020-12-29 北京百度网讯科技有限公司 图像数据库的更新方法、装置、电子设备和介质
CN112148907A (zh) * 2020-10-23 2020-12-29 北京百度网讯科技有限公司 图像数据库的更新方法、装置、电子设备和介质
CN112488985A (zh) * 2019-09-11 2021-03-12 上海高德威智能交通系统有限公司 图像质量确定方法、装置及设备
CN112686847A (zh) * 2020-12-23 2021-04-20 平安银行股份有限公司 身份证图像拍摄质量评价方法、装置、计算机设备和介质
CN113095672A (zh) * 2021-04-09 2021-07-09 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
CN113705496A (zh) * 2021-08-31 2021-11-26 深圳市酷开网络科技股份有限公司 海报选择方法、装置、设备及存储介质

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109145717B (zh) * 2018-06-30 2021-05-11 东南大学 一种在线学习的人脸识别方法
CN111126121B (zh) * 2018-11-01 2023-04-04 百度在线网络技术(北京)有限公司 人脸识别模型的调整方法、装置、设备及存储介质
CN109544523B (zh) * 2018-11-14 2021-01-01 北京智芯原动科技有限公司 基于多属性人脸比对的人脸图像质量评价方法及装置
CN109753917A (zh) * 2018-12-29 2019-05-14 中国科学院重庆绿色智能技术研究院 人脸质量寻优方法、系统、计算机可读存储介质及设备
CN110427888A (zh) * 2019-08-05 2019-11-08 北京深醒科技有限公司 一种基于特征聚类的人脸质量评估方法
CN112989869B (zh) * 2019-12-02 2024-05-07 深圳云天励飞技术有限公司 人脸质量检测模型的优化方法、装置、设备及存储介质
CN111709906A (zh) * 2020-04-13 2020-09-25 北京深睿博联科技有限责任公司 医学图像的质量评估方法和装置
CN117372405A (zh) * 2023-10-31 2024-01-09 神州通立电梯有限公司 人脸图像质量评估方法、装置、存储介质及设备

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20070174272A1 (en) * 2005-06-24 2007-07-26 International Business Machines Corporation Facial Recognition in Groups
US20080080745A1 (en) * 2005-05-09 2008-04-03 Vincent Vanhoucke Computer-Implemented Method for Performing Similarity Searches
CN106096538A (zh) * 2016-06-08 2016-11-09 中国科学院自动化研究所 基于定序神经网络模型的人脸识别方法及装置
CN106203333A (zh) * 2016-07-08 2016-12-07 乐视控股(北京)有限公司 人脸识别方法及系统
CN106815566A (zh) * 2016-12-29 2017-06-09 天津中科智能识别产业技术研究院有限公司 一种基于多任务卷积神经网络的人脸检索方法

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8620092B2 (en) * 2010-03-04 2013-12-31 Hewlett-Packard Development Company, L.P. Determining similarity of two images
EP2676224B1 (en) * 2011-02-18 2021-05-26 iOmniscient Pty Ltd Image quality assessment
JP5988577B2 (ja) * 2011-12-28 2016-09-07 キヤノン株式会社 画像符号化装置、画像符号化方法及びプログラム
CN104636730B (zh) * 2015-02-10 2018-09-04 北京信息科技大学 人脸验证的方法和装置
CN104866864A (zh) * 2015-05-07 2015-08-26 天津大学 一种用于立体图像质量客观评价的极端学习机
CN105608700B (zh) * 2015-12-24 2019-12-17 广州视源电子科技股份有限公司 照片的筛选方法及系统
CN106503691B (zh) * 2016-11-10 2019-12-20 广州视源电子科技股份有限公司 一种人脸图片的身份标注方法和装置

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080080745A1 (en) * 2005-05-09 2008-04-03 Vincent Vanhoucke Computer-Implemented Method for Performing Similarity Searches
US20070174272A1 (en) * 2005-06-24 2007-07-26 International Business Machines Corporation Facial Recognition in Groups
CN106096538A (zh) * 2016-06-08 2016-11-09 中国科学院自动化研究所 基于定序神经网络模型的人脸识别方法及装置
CN106203333A (zh) * 2016-07-08 2016-12-07 乐视控股(北京)有限公司 人脸识别方法及系统
CN106815566A (zh) * 2016-12-29 2017-06-09 天津中科智能识别产业技术研究院有限公司 一种基于多任务卷积神经网络的人脸检索方法

Cited By (21)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109977815A (zh) * 2019-03-13 2019-07-05 上海商汤智能科技有限公司 图像质量评价方法及装置、电子设备、存储介质
CN111797845A (zh) * 2019-03-19 2020-10-20 北京沃东天骏信息技术有限公司 图片处理方法、装置、存储介质及电子设备
CN112488985A (zh) * 2019-09-11 2021-03-12 上海高德威智能交通系统有限公司 图像质量确定方法、装置及设备
CN110866437A (zh) * 2019-09-23 2020-03-06 平安科技(深圳)有限公司 颜值判定模型优化方法、装置、电子设备及存储介质
CN111191584A (zh) * 2019-12-30 2020-05-22 电信科学技术第十研究所有限公司 一种人脸识别方法及装置
CN111191584B (zh) * 2019-12-30 2024-02-09 电信科学技术第十研究所有限公司 一种人脸识别方法及装置
CN111311581A (zh) * 2020-02-20 2020-06-19 杭州涂鸦信息技术有限公司 基于光照的图片评分方法及其系统和装置
CN111382681A (zh) * 2020-02-28 2020-07-07 浙江大华技术股份有限公司 一种人脸注册方法、装置及存储介质
CN111382681B (zh) * 2020-02-28 2023-11-14 浙江大华技术股份有限公司 一种人脸注册方法、装置及存储介质
CN111382693A (zh) * 2020-03-05 2020-07-07 北京迈格威科技有限公司 图像质量的确定方法、装置、电子设备和计算机可读介质
CN111814570A (zh) * 2020-06-12 2020-10-23 深圳禾思众成科技有限公司 一种基于动态阈值的人脸识别方法、系统及存储介质
CN111814570B (zh) * 2020-06-12 2024-04-30 深圳禾思众成科技有限公司 一种基于动态阈值的人脸识别方法、系统及存储介质
CN111966852A (zh) * 2020-06-28 2020-11-20 北京百度网讯科技有限公司 基于人脸的虚拟整容的方法和装置
CN111966852B (zh) * 2020-06-28 2024-04-09 北京百度网讯科技有限公司 基于人脸的虚拟整容的方法和装置
CN112148907A (zh) * 2020-10-23 2020-12-29 北京百度网讯科技有限公司 图像数据库的更新方法、装置、电子设备和介质
CN112148908A (zh) * 2020-10-23 2020-12-29 北京百度网讯科技有限公司 图像数据库的更新方法、装置、电子设备和介质
CN112686847A (zh) * 2020-12-23 2021-04-20 平安银行股份有限公司 身份证图像拍摄质量评价方法、装置、计算机设备和介质
CN112686847B (zh) * 2020-12-23 2024-05-14 平安银行股份有限公司 身份证图像拍摄质量评价方法、装置、计算机设备和介质
CN113095672A (zh) * 2021-04-09 2021-07-09 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
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
CN113705496A (zh) * 2021-08-31 2021-11-26 深圳市酷开网络科技股份有限公司 海报选择方法、装置、设备及存储介质

Also Published As

Publication number Publication date
CN107609493B (zh) 2021-04-13
CN107609493A (zh) 2018-01-19

Similar Documents

Publication Publication Date Title
WO2019037346A1 (zh) 优化人脸图片质量评价模型的方法及装置
Hofbauer et al. A ground truth for iris segmentation
KR101725651B1 (ko) 식별 장치 및 식별 장치의 제어 방법
WO2020155939A1 (zh) 图像识别方法、装置、存储介质和处理器
WO2019127451A1 (zh) 图像识别方法及云端系统
US11804071B2 (en) Method for selecting images in video of faces in the wild
JP5633080B2 (ja) 属性値推定装置、属性値推定方法、プログラム及び記録媒体
US8165354B1 (en) Face recognition with discriminative face alignment
KR102284096B1 (ko) 중요도 맵을 이용한 지능형 주관적 화질 평가 시스템, 방법, 및 상기 방법을 실행시키기 위한 컴퓨터 판독 가능한 프로그램을 기록한 기록 매체
CN106446754A (zh) 图像识别方法、度量学习方法、图像来源识别方法及装置
US20070174272A1 (en) Facial Recognition in Groups
EP3680794A1 (en) Device and method for user authentication on basis of iris recognition
US20160253549A1 (en) Estimating personal information from facial features
KR20170006355A (ko) 모션벡터 및 특징벡터 기반 위조 얼굴 검출 방법 및 장치
WO2021000832A1 (zh) 匹配人脸的方法、装置、计算机设备和存储介质
WO2009158700A1 (en) Assessing biometric sample quality using wavelets and a boosted classifier
Gupta et al. Identification of age, gender, & race SMT (scare, marks, tattoos) from unconstrained facial images using statistical techniques
JP5812505B2 (ja) マルチモーダル情報に基づく人口学的分析方法及びシステム
WO2021017316A1 (zh) 基于残差网络的信息识别方法、装置和计算机设备
WO2020019457A1 (zh) 用户指令匹配方法、装置、计算机设备及存储介质
CN107480628B (zh) 一种人脸识别方法及装置
WO2015131710A1 (zh) 人眼定位方法及装置
Kleinberg et al. A study of quantitative comparisons of photographs and video images based on landmark derived feature vectors
Tavakolian et al. Face recognition under occlusion for user authentication and invigilation in remotely distributed online assessments
Paone Liberating the biometric menagerie through score normalization improvements

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: 17922228

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: 17922228

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 16/10/2020)

122 Ep: pct application non-entry in european phase

Ref document number: 17922228

Country of ref document: EP

Kind code of ref document: A1