WO2020252917A1 - 一种模糊人脸图像识别方法、装置、终端设备及介质 - Google Patents

一种模糊人脸图像识别方法、装置、终端设备及介质 Download PDF

Info

Publication number
WO2020252917A1
WO2020252917A1 PCT/CN2019/103159 CN2019103159W WO2020252917A1 WO 2020252917 A1 WO2020252917 A1 WO 2020252917A1 CN 2019103159 W CN2019103159 W CN 2019103159W WO 2020252917 A1 WO2020252917 A1 WO 2020252917A1
Authority
WO
WIPO (PCT)
Prior art keywords
face
image
blurred
fuzzy
preset
Prior art date
Application number
PCT/CN2019/103159
Other languages
English (en)
French (fr)
Inventor
徐玲玲
Original Assignee
平安科技(深圳)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 平安科技(深圳)有限公司 filed Critical 平安科技(深圳)有限公司
Publication of WO2020252917A1 publication Critical patent/WO2020252917A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/21Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16Human faces, e.g. facial parts, sketches or expressions

Definitions

  • This application belongs to the field of data processing technology, and in particular relates to a fuzzy face image recognition method, device, terminal equipment and medium.
  • the final face image will be blurred.
  • the face image is applied in the next step, it will greatly affect the effect of the actual application. For example, when it is used as the sample data for face recognition and attendance, it will greatly reduce the reliability of face matching, so there is a need for a fuzzy face image that can be identified
  • the method to realize the recognition and screening of fuzzy face images are described in detail below.
  • fuzzy image recognition methods there are some fuzzy image recognition methods in the prior art, but they are all processed and recognized for fuzzy images with lower resolution.
  • these fuzzy image recognition is not designed with the face itself, it makes the fuzzy recognition of the face At times, it is extremely susceptible to the influence of the non-face area in the image, resulting in inaccurate recognition. Therefore, the prior art urgently needs a fuzzy face image recognition method that can be adapted to various types.
  • the embodiments of the present application provide a fuzzy face image recognition method and terminal device to solve the problem that the prior art cannot adapt to various different types of fuzzy face image recognition, so that it is compatible with fuzzy face image recognition.
  • the first aspect of the embodiments of the present application provides a fuzzy face image recognition method, including:
  • the recognition model obtained by training with a variety of fuzzy types of fuzzy face region image samples is used to identify whether the image is a blurred image;
  • the image of the face area to be processed is a blurred image, perform face recognition on the image to be processed based on a preset face recognition model, and determine the probability that the image to be processed contains a face;
  • the probability that the image to be processed contains a human face is less than a preset face probability threshold, it is determined that the image to be processed is a blurred human face image.
  • a second aspect of the embodiments of the present application provides a fuzzy face image recognition device, including:
  • the face cropping module is used to crop the face contour of the image to be processed and adjust it to a preset image size to obtain the corresponding image of the face area to be processed;
  • the fuzzy image recognition module is used to input the face area image to be processed into a pre-trained fuzzy recognition model for processing, and to identify whether the face area image to be processed is a blurred image, wherein the fuzzy recognition model is a pre-trained fuzzy recognition model.
  • a recognition model obtained by training based on clear face region image samples and multiple fuzzy types of blurred face region image samples is used to identify whether the image is a blurred image;
  • the face recognition module is configured to, if the image of the face area to be processed is a blurred image, perform face recognition on the image to be processed based on a preset face recognition model, and determine that the image to be processed contains a human face Probability
  • the fuzzy face recognition module is used to determine that the image to be processed is a blurred face image if the probability that the image to be processed contains a human face is less than a preset face probability threshold.
  • a third aspect of the embodiments of the present application provides a terminal device, including a memory and a processor.
  • the memory stores computer-readable instructions that can run on the processor.
  • the processor executes the computer The following steps are implemented when reading instructions:
  • the recognition model obtained by training with a variety of fuzzy types of fuzzy face region image samples is used to identify whether the image is a blurred image;
  • the image of the face area to be processed is a blurred image, perform face recognition on the image to be processed based on a preset face recognition model, and determine the probability that the image to be processed contains a face;
  • the probability that the image to be processed contains a human face is less than a preset face probability threshold, it is determined that the image to be processed is a blurred human face image.
  • a fourth aspect of the embodiments of the present application provides a computer-readable storage medium that stores computer-readable instructions, wherein the computer-readable instructions are implemented when executed by at least one processor The following steps:
  • the recognition model obtained by training with a variety of fuzzy types of fuzzy face region image samples is used to identify whether the image is a blurred image;
  • the image of the face area to be processed is a blurred image, perform face recognition on the image to be processed based on a preset face recognition model, and determine the probability that the image to be processed contains a face;
  • the probability that the image to be processed contains a human face is less than a preset face probability threshold, it is determined that the image to be processed is a blurred human face image.
  • the embodiment of the application has the following beneficial effects: on the one hand, the face contour cropping and image size adjustment are performed on the image to be processed in advance, which not only preserves the key facial features, but also makes the judgment of the facial image blur It can focus on the face itself, reducing the interference of the sharpness of the non-face area. On the other hand, it is based on the actual clear face area image sample and the fuzzy face area image sample training constructed by the fuzzy recognition.
  • Model to perform fuzzy recognition on the image of the face area to be processed, realize the compatibility and accurate recognition of multiple types of facial blur, accurately determine whether the face area to be processed is blurred, and at the same time, in order to prevent face contour recognition and fuzzy recognition model recognition
  • the possible interference in the system leads to an increase in the recognition error.
  • the embodiment of the present application further performs a secondary check on whether the image to be processed that is judged to be a blurred image contains a human face, and when the probability of containing a human face is small, That is, when the accuracy of face recognition is low due to various blur factors, the image to be processed is finally determined to be a blurred face image, which greatly improves the final compatibility and accuracy of different types of blurred face image recognition .
  • FIG. 1 is a schematic diagram of the implementation process of the fuzzy face image recognition method provided by Embodiment 1 of the present application;
  • FIG. 2 is a schematic diagram of the implementation process of the fuzzy face image recognition method provided in the second embodiment of the present application
  • FIG. 3 is a schematic diagram of the implementation process of the fuzzy face image recognition method provided by Embodiment 3 of the present application;
  • FIG. 4 is a schematic diagram of the implementation process of the fuzzy face image recognition method provided by the fourth embodiment of the present application.
  • FIG. 5 is a schematic structural diagram of a fuzzy face image recognition device provided by Embodiment 5 of the present application.
  • FIG. 6 is a schematic diagram of a terminal device provided in Embodiment 6 of the present application.
  • various different types of blurred face image samples are first collected in advance, and the fuzzy recognition model is trained based on these samples.
  • a fuzzy recognition model that can perform fuzzy image recognition is obtained.
  • pre-processing the image to perform face contour cropping and image size adjustment it not only retains the key facial features, but also makes the facial image blurred.
  • the judgment can be focused on the face itself, reducing the interference of the definition of the non-face area.
  • the fuzzy recognition of the processed face area image based on the trained fuzzy recognition model is realized to realize the recognition of a variety of faces.
  • Compatibility of the blur type is accurately recognized, and the face area to be processed is accurately judged whether the face area to be processed is blurred.
  • the embodiment of the present application further Perform a secondary check on whether the image to be processed of the blurred image contains a face, and only when the probability of containing a face is small, that is, when the accuracy of face recognition is low due to various blur factors, the final judgment is made.
  • the image to be processed is a blurred face image, which greatly improves the compatibility and accuracy of final recognition of different types of blurred face images. The details are as follows:
  • Fig. 1 shows the implementation flow chart of the fuzzy face image recognition method provided in the first embodiment of the present application, and the details are as follows:
  • S101 Perform face contour cropping on an image to be processed and adjust it to a preset image size to obtain a corresponding face region image to be processed.
  • the embodiment of the present application first performs face contour cropping on the image to be processed, that is, recognizes the face area in the image to be processed, and It is cropped into an independent face region image to be processed.
  • face contour cropping on the image to be processed, that is, recognizes the face area in the image to be processed, and It is cropped into an independent face region image to be processed.
  • the embodiment of the present application will uniformly scale the size of the extracted face area and adjust it to the preset size. Set the image size, where the specific image size can be set by the technician according to actual needs, preferably, it can be set to 112 ⁇ 112 pixels.
  • the embodiment of the present application itself needs to recognize a blurred face image
  • the face area contained in the image to be processed is also very likely to be blurred, making it difficult to accurately recognize the face area itself. If the requirements for face recognition are too high, it is very likely that some fuzzy face regions will not be recognized, so that it is impossible to perform face contour cropping on the processed image. Therefore, when performing face contour cropping in the embodiment of this application, Use some face recognition methods with lower accuracy requirements for face recognition search and cropping, including but not limited to face contour positioning algorithms for face contour search, and recognize the image of the area that meets the face contour as the face area Make a ruling, or reduce the threshold of face recognition judgment in some traditional face recognition algorithms, so as to improve the ability to recognize blurred faces.
  • multiple face regions can be obtained during face recognition, such as objects similar to the contour of a human face and other human faces other than the target face that were accidentally captured during shooting.
  • face area recognition when performing face area recognition, only the area meeting the requirements of face recognition and having the most pixels is recognized as the face area, and subsequent cropping is performed.
  • S102 Input the face area image to be processed into a pre-trained fuzzy recognition model for processing, and identify whether the face area image to be processed is a blurred image, where the fuzzy recognition model is based on a clear face area image sample and multiple A recognition model obtained by training a fuzzy type of fuzzy face region image sample is used to identify whether the image is a blurred image.
  • fuzzy recognition models include, but are not limited to, binary classification models and neural network models, which are specifically selected by the technicians according to actual needs, and are not limited here.
  • specific model training methods can also be set by the technicians.
  • the embodiment of the present application inputs the cropped face region image to be processed into the fuzzy recognition model, and the model is used for processing to identify whether the face region to be processed is a blurred image.
  • S103 If the image of the face area to be processed is a blurred image, perform face recognition on the image to be processed based on a preset face recognition model, and determine the probability that the image to be processed contains a face.
  • the face image is picked out. It is precisely based on the above-mentioned practical application requirements that when the embodiment of the present application recognizes the face area image to be processed as a blurred image, it will not directly determine it as a blurred face image, but will perform a secondary check on it. In order to ensure that the final recognition result can meet the needs of practical applications, specifically: considering that whatever fuzzy interference factors will eventually cause the face area itself to become blurred, the embodiment of this application will determine when the second inspection Perform face recognition for the face region image to be processed in the blurred image to obtain the probability that it contains a face.
  • the probability is less than the preset face probability threshold, it means that the face region image in the face region image to be processed has a higher degree of facial blur . It is difficult to recognize the human face normally, so the embodiment of the present application will directly determine that the human face is blurred, that is, the image to be processed is a blurred human face image.
  • the specific face recognition model used is not limited here, but unlike step S101, it is necessary to perform relatively accurate face search and matching on the image of the face area to be processed to determine the degree of blur. It is necessary to select some high-precision face recognition methods for face recognition, including but not limited to some face recognition models based on deep learning model training, or face recognition models based on facial feature point matching.
  • the specific value of the face probability threshold can be set by the technicians themselves, or obtained by processing with reference to the fourth embodiment of the present application, which is not limited here.
  • various different types of fuzzy face image samples are collected in advance, and the fuzzy recognition model is trained based on these samples, so as to obtain a fuzzy recognition model that can perform fuzzy image recognition.
  • the fuzzy recognition model is trained based on these samples, so as to obtain a fuzzy recognition model that can perform fuzzy image recognition.
  • the fuzzy recognition model is trained based on these samples, so as to obtain a fuzzy recognition model that can perform fuzzy image recognition.
  • the image to be processed for face contour cropping and image size adjustment not only the key facial features are retained, but also the judgment of the facial image blur can be focused on the face itself, reducing the clarity interference of the non-face area
  • the fuzzy recognition of the image of the face area to be processed based on the trained fuzzy recognition model the compatibility and accurate recognition of multiple face blur types are realized, and the face area to be processed is accurately judged whether the face area is blurred.
  • the embodiment of the present application further determines Perform a secondary check on whether the processed image of the blurred image contains a human face, and only when the probability of containing a human face is small, that is, when the degree of facial blur is high and it is difficult to recognize the human face normally, the final The image to be processed is determined to be a blurred face image, thereby greatly improving the compatibility and accuracy of the final recognition of different types of blurred face images.
  • the second embodiment of the present application includes:
  • S201 Obtain a sample of a clear face image and a sample of a blurred face image of multiple blur types.
  • the way to obtain the face image sample can be either crawling from the website or shooting by the technician. It is not limited here, but it should be ensured that the obtained blurred face image sample contains the type of light blur and motion blur. Type and resolution blur type of the fuzzy face image samples to ensure the effectiveness of the subsequently trained model. At the same time, the number of the three types of fuzzy face image samples should be approximately the same or similar.
  • the blurred face image sample includes:
  • S301 Acquire a first preset number of blurred face image samples of a light blur type obtained by image collection of a face under different light conditions.
  • different light intensities can be adjusted and multiple light types can be used to change the collection environment and perform face image collection.
  • S302 Acquire a second preset number of blurred facial image samples of a motion blur type taken by the camera at various relative movement speeds different from the human body.
  • different relative movement speeds can also be adjusted to collect face images, so as to obtain blurred face image samples with different blur degrees corresponding to the motion blur types.
  • S303 Obtain a clear face image sample, and perform pixel blurring processing on the clear face image sample to obtain a third preset number of blurred face image samples of a resolution blur type.
  • the pixel blur processing methods here include but are not limited to, for example, the use of mean blur, median blur, Gaussian blur, bilateral blur and zoom blur to blur clear face image samples to obtain the corresponding resolution blur type of blurred face image samples .
  • the number of the three types of fuzzy face image samples in the embodiment of the present application should be approximately close or the same.
  • the first preset number, the second preset number, and the third The value of the preset quantity can be set by the technician.
  • the embodiment of the present application further includes: crawling a preset website to obtain a fourth preset number of blurred face image samples.
  • the fourth preset number, the first preset number, the second preset number, and the third preset number decrease sequentially.
  • the fuzzy face images Due to the large number of samples required, the difficulty of all shooting/processing and the insufficient randomness of the samples, it is difficult to guarantee the final effect. Therefore, it is necessary to crawl the fuzzy face images that may appear in real life, but in actual situations, it can be crawled.
  • the fuzzy type contained in the obtained fuzzy face data is difficult to control. According to the analysis of the image data that can be crawled on the actual website, it is found that the number of various types of fuzzy face images is: resolution blur type >Motion blur type>Light blur type. Therefore, based on actual crawling, in order to balance the number of three types of blurred face images as much as possible, in the embodiment of this application, the first preset number and the second preset number The quantity and the third preset quantity have changed. The fourth preset quantity> the first preset quantity> the second preset quantity> the third preset quantity needs to be met. The specific quantity needs to be determined by the technician The actual crawling situation is set.
  • S202 Perform face contour cropping on the clear face image sample and the blurred face image sample and adjust them to a preset image size to obtain corresponding clear face area image samples and blurred face area image samples.
  • both clear face image samples and blurred face image samples are performed.
  • Face contour cropping, and in the embodiments of this application, fuzzy face recognition is required instead of fine face recognition or feature extraction. Therefore, too much face area data is not required, so when selecting the size
  • the embodiment of the present application preferably selects a smaller size to increase the speed of model training and the recognition speed of the blurred face image of the image to be processed in the first embodiment of the present application as much as possible on the basis of ensuring fuzzy recognition.
  • the specific face contour cutting method can refer to the related description in the first embodiment of the present application, which will not be repeated here.
  • S203 Train a preset two-class classification model based on the clear face area image sample and the blurred face area image sample, and calculate the cross-entropy loss value of the classification result.
  • the face area image sample includes a clear face area image sample and a blurred face area image sample.
  • the specific two-classification model used is not limited here, including but not limited to, for example, the two-classification model based on logistic regression model and the two-classification model based on machine learning, etc., which can be set by technicians according to actual needs.
  • the accuracy of the inspection will be directly affected by the accuracy of the adopted face recognition model and the adopted face probability
  • the validity of the threshold is affected. If the technician directly sets a fixed face probability threshold for judgment, the accuracy of the inspection is easily affected by the subjectivity of the technician, which makes the accuracy difficult to guarantee. Therefore, in order to improve the inspection Effectiveness, as shown in Figure 4, includes:
  • S401 Perform face recognition on the clear face image sample and the blurred face image sample based on the face recognition model, and obtain the probability that each clear face image sample and the blurred face image sample respectively contain a face.
  • S402 Determine a preset face probability threshold based on the probability that each clear face image sample and the blurred face image sample respectively contain a face.
  • the actual selection of the first embodiment of this application will be used in this application.
  • Face recognition model to identify the actual face image samples obtained, and compare the recognition results with the actual labels of each face image sample, and select the ones that can make the two-classification effect the best.
  • the embodiment of the present application uses 70% as the face probability threshold in the first embodiment of the present application.
  • the face probability threshold is calculated and selected based on the actual face recognition model and face image samples, which can avoid the recognition inaccuracy caused by the recognition accuracy of different face recognition models to this application as much as possible.
  • the embodiments of this application can well determine the appropriate threshold according to the recognition ability of the face recognition model on the actual sample, so that the present application In the embodiment, a pair of clear face recognition capabilities realize adaptive optimization.
  • FIG. 5 shows a structural block diagram of a fuzzy face image recognition device provided in an embodiment of the present application.
  • the fuzzy face image recognition device illustrated in Fig. 5 may be the subject of execution of the fuzzy face image recognition method provided in the first embodiment.
  • the fuzzy face image recognition device includes:
  • the face cropping module 51 is used to crop the face contour of the image to be processed and adjust it to a preset image size to obtain the corresponding face region image to be processed.
  • the fuzzy image recognition module 52 is used to input the face area image to be processed into a pre-trained fuzzy recognition model for processing, and to identify whether the face area image to be processed is a blurred image, where the fuzzy recognition model is A recognition model trained in advance based on clear face region image samples and multiple blur types of blurred face region image samples is used to identify whether the image is a blurred image.
  • the face recognition module 53 is configured to, if the face region image to be processed is a blurred image, perform face recognition on the image to be processed based on a preset face recognition model, and determine that the image to be processed contains a face The probability.
  • the fuzzy face recognition module 54 is configured to determine that the image to be processed is a blurred face image if the probability that the image to be processed contains a human face is less than a preset face probability threshold.
  • the fuzzy face image recognition device further includes:
  • the sample acquisition module is used to acquire clear face image samples and blurred face image samples of various blur types.
  • the sample cropping module is used to crop the human face contour of the clear human face image sample and the blurred human face image sample and adjust it to the preset image size to obtain the corresponding clear human face region image sample and The image sample of the blurred face area.
  • the model training module is configured to train a preset binary classification model based on the clear face area image sample and the blurred face area image sample, and calculate the loss value of the cross entropy of the classification result.
  • the iterative training module is used to update and train the binary classification model based on the gradient descent method if the loss value is greater than the preset loss threshold, until the loss value is less than or equal to the preset loss threshold or the number of iterations reaches the preset Set the frequency threshold, complete the training, and obtain the fuzzy recognition model.
  • sample acquisition module includes:
  • sample acquisition module also includes:
  • model training module includes
  • the loss value of the cross entropy of the classification result is calculated based on the following formula (1):
  • the face area image sample includes a clear face area image sample and a blurred face area image sample.
  • the fuzzy face image recognition device further includes:
  • the size of the sequence number of each step in the foregoing embodiment does not mean the order of execution.
  • the execution sequence of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiment of the present application.
  • the terms “first”, “second”, etc. are used in the text in some embodiments of the present application to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another.
  • the first table may be named the second table, and similarly, the second table may be named the first table without departing from the scope of the various described embodiments.
  • the first form and the second form are both forms, but they are not the same form.
  • Fig. 6 is a schematic diagram of a terminal device provided by an embodiment of the present application.
  • the terminal device 6 of this embodiment includes a processor 60 and a memory 61.
  • the memory 61 stores computer readable instructions 62 that can run on the processor 60.
  • the steps in the foregoing embodiments of the fuzzy face image recognition method are implemented, such as steps 101 to 105 shown in FIG. 1.
  • the processor 60 executes the computer-readable instructions 62
  • the functions of the modules/units in the foregoing device embodiments, such as the functions of the modules 51 to 54 shown in FIG. 5, are realized.
  • the terminal device 6 may be a computing device such as a desktop computer, a notebook, a palmtop computer, and a cloud server.
  • the terminal device may include, but is not limited to, a processor 60 and a memory 61.
  • FIG. 6 is only an example of the terminal device 6 and does not constitute a limitation on the terminal device 6. It may include more or less components than shown in the figure, or a combination of certain components, or different components.
  • the terminal device may also include an input sending device, a network access device, a bus, and the like.
  • the so-called processor 60 may be a central processing unit (Central Processing Unit, CPU), other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), Ready-made programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.
  • the general-purpose processor may be a microprocessor or the processor may also be any conventional processor or the like.
  • the memory 61 may be an internal storage unit of the terminal device 6, such as a hard disk or a memory of the terminal device 6.
  • the memory 61 may also be an external storage device of the terminal device 6, for example, a plug-in hard disk equipped on the terminal device 6, a smart memory card (Smart Media Card, SMC), or a Secure Digital (SD) Card, Flash Card, etc.
  • the memory 61 may also include both an internal storage unit of the terminal device 6 and an external storage device.
  • the memory 61 is used to store the computer readable instructions and other programs and data required by the terminal device.
  • the memory 61 can also be used to temporarily store data that has been sent or will be sent.
  • the functional units in each embodiment of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units may be integrated into one unit.
  • the above-mentioned integrated unit can be implemented in the form of hardware or software functional unit.
  • the integrated module/unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium.
  • this application implements all or part of the procedures in the above-mentioned embodiments and methods, and can also be completed by instructing relevant hardware through computer-readable instructions, and the computer-readable instructions can be stored in a computer-readable storage medium.
  • the computer-readable instruction is executed by the processor, the steps of the foregoing method embodiments can be implemented.
  • Non-volatile memory may include read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory.
  • ROM read only memory
  • PROM programmable ROM
  • EPROM electrically programmable ROM
  • EEPROM electrically erasable programmable ROM
  • Volatile memory may include random access memory (RAM) or external cache memory.
  • RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain Channel (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Human Computer Interaction (AREA)
  • Multimedia (AREA)
  • Image Analysis (AREA)

Abstract

一种模糊人脸图像识别方法、装置、终端设备及介质,适用于数据处理技术领域,该方法包括:对待处理图像进行人脸轮廓裁剪并调整至预设图像尺寸,得到对应的待处理人脸区域图像(S101);将待处理人脸区域图像输入至预先训练好的模糊识别模型中进行处理,识别待处理人脸区域图像是否为模糊图像(S102);若待处理人脸区域图像为模糊图像,则基于预设的人脸识别模型对待处理图像进行人脸识别,判断待处理图像包含人脸的概率(S103);若待处理图像包含人脸的概率小于预设人脸概率阈值,则判定待处理图像为模糊人脸图像(S104)。该方法极大地提升了最终对不同类型的模糊人脸图像识别的兼容性和准确率。

Description

一种模糊人脸图像识别方法、装置、终端设备及介质
本申请要求于2019年06月20日提交中国专利局、申请号为201910536758.X、发明名称为“一种模糊人脸图像识别方法、装置及终端设备”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。
技术领域
本申请属于数据处理技术领域,尤其涉及一种模糊人脸图像识别方法、装置、终端设备及介质。
背景技术
在对人脸图像的采集、传输和保存的过程中,由于拍摄设备或用户的运动、环境的光线以及拍摄设备的配置等因素影响,会造成最终得到的人脸图像模糊,而在使用这些模糊人脸图像进行下一步应用时,会极大地影响实际应用的效果,如作为人脸识别考勤的样本数据时,会极大地降低人脸匹配的可靠性,因此需要一种可以识别模糊人脸图像的方法,以实现对模糊人脸图像的识别筛选。
现有技术中存在一些模糊图像识别的方法,但都是针对分辨率较低的模糊图像进行处理识别,一方面如上所述,造成人脸图像模糊的可能因素很多,因此实际应用中可能存在的模糊人脸图像类型也有多种,例如物体运动导致的运动模糊、光线不佳导致的光线模糊以及由于拍摄分辨率较低导致的分辨率模糊,由于现有技术仅能对分辨率模糊进行处理,对于其他类型的模糊图像识别效果并不友好,从而导致最终实际的识别效果并不理想,另一方面,由于这些模糊图像识别并没有结合人脸本身进行设计,使得在对人脸进行模糊识别的时候极易受到图像中非人脸区域的影响,从而造成识别不准确的情况,因此,现有技术急需一种可以适应于各种不同类型的模糊人脸图像识别方法。
技术问题
有鉴于此,本申请实施例提供了一种模糊人脸图像识别方法及终端设备,以解决现有技术中无法适应于各种不同类型的模糊人脸图像识别,使得对模糊人脸图像识别兼容性和准确率较低的问题。
技术解决方案
本申请实施例的第一方面提供了一种模糊人脸图像识别方法,包括:
对待处理图像进行人脸轮廓裁剪并调整至预设图像尺寸,得到对应的待处理人脸区域图像;
将所述待处理人脸区域图像输入至预先训练好的模糊识别模型中进行处理,识别所述待处理人脸区域图像是否为模糊图像,其中,模糊识别模型为预先基于清晰人脸区域图像样本和多种模糊类型的模糊人脸区域图像样本进行训练得到的识别模型,用于识别图像是否为模糊图像;
若所述待处理人脸区域图像为模糊图像,则基于预设的人脸识别模型对所述待处理图像进行人脸识别,判断所述待处理图像包含人脸的概率;
若所述待处理图像包含人脸的概率小于预设人脸概率阈值,则判定所述待处理图像为模糊人脸图像。
本申请实施例的第二方面提供了一种模糊人脸图像识别装置,包括:
人脸裁剪模块,用于对待处理图像进行人脸轮廓裁剪并调整至预设图像尺寸,得到对应的待处理人脸区域图像;
模糊图像识别模块,用于将所述待处理人脸区域图像输入至预先训练好的模糊识别模型中进行处理,识别所述待处理人脸区域图像是否为模糊图像,其中,模糊识别模型为预先基于清晰人脸区域图像样本和多种模糊类型的模糊人脸区域图像样本进行训练得到的识别模型,用于识别图像是否为模糊图像;
人脸识别模块,用于若所述待处理人脸区域图像为模糊图像,则基于预设的人脸识别模型对所述待处理图像进行人脸识别,判断所述待处理图像包含人脸的概率;
模糊人脸识别模块,用于若所述待处理图像包含人脸的概率小于预设人脸概率阈值,则判定所述待处理图像为模糊人脸图像。
本申请实施例的第三方面提供了一种终端设备,包括存储器、处理器,所述存储器上存储有可在所述处理器上运行的计算机可读指令,所述处理器执行所述计算机可读指令时实现如下步骤:
对待处理图像进行人脸轮廓裁剪并调整至预设图像尺寸,得到对应的待处理人脸区域图像;
将所述待处理人脸区域图像输入至预先训练好的模糊识别模型中进行处理,识别所述待处理人脸区域图像是否为模糊图像,其中,模糊识别模型为预先基于清晰人脸区域图像样本和多种模糊类型的模糊人脸区域图像样本进行训练得到的识别模型,用于识别图像是否为模糊图像;
若所述待处理人脸区域图像为模糊图像,则基于预设的人脸识别模型对所述待处理图像进行人脸识别,判断所述待处理图像包含人脸的概率;
若所述待处理图像包含人脸的概率小于预设人脸概率阈值,则判定所述待处理图像为模糊人脸图像。
本申请实施例的第四方面提供了一种计算机可读存储介质,所述计算机可读存储介质存储有计算机可读指令,其特征在于,所述计算机可读指令被至少一个处理器执行时实现如下步骤:
对待处理图像进行人脸轮廓裁剪并调整至预设图像尺寸,得到对应的待处理人脸区域图像;
将所述待处理人脸区域图像输入至预先训练好的模糊识别模型中进行处理,识别所述待处理人脸区域图像是否为模糊图像,其中,模糊识别模型为预先基于清晰人脸区域图像样本和多种模糊类型的模糊人脸区域图像样本进行训练得到的识别模型,用于识别图像是否为模糊图像;
若所述待处理人脸区域图像为模糊图像,则基于预设的人脸识别模型对所述待处理图像进行人脸识别,判断所述待处理图像包含人脸的概率;
若所述待处理图像包含人脸的概率小于预设人脸概率阈值,则判定所述待处理图像为模糊人脸图像。
有益效果
本申请实施例与现有技术相比存在的有益效果是:一方面通过预先对待处理图像进行人脸轮廓裁剪和图像尺寸调整,既保留了关键人脸特征,又使得对人脸图像模糊的判断能集中于人脸本身,减小了非人脸区域的清晰度干扰,另一方面,再通过基于实际清晰人脸区域图像样本和多种模糊类型的模糊人脸区域图像样本训练构建的模糊识别模型,对待处理人脸区域图像进行模糊识别,实现了对多种人脸模糊类型的兼容性准确识别,准确判断待处理人脸区域是否模糊,同时,为了防止人脸轮廓识别和模糊识别模型识别中可能存在的干扰导致识别误差增大,本申请实施例还进一步地对判断为模糊图像的待处理图像进行是否包含人脸的二次校验,并在包含人脸概率较小的情况下,即各种模糊因素导致人脸识别准确率较低的情况下,才最终判定待处理图像为模糊人脸图像,从而极大地提升了最终对不同类型的模糊人脸图像识别的兼容性和准确率。
附图说明
为了更清楚地说明本申请实施例中的技术方案,下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍,显而易见地,下面描述中的附图仅仅是本申请的一些实施例,对于本领域普通技术人员来讲,在不付出创造性劳动性的前提下,还可以根据这些附图获得其他的附图。
图1是本申请实施例一提供的模糊人脸图像识别方法的实现流程示意图;
图2是本申请实施例二提供的模糊人脸图像识别方法的实现流程示意图;
图3是本申请实施例三提供的模糊人脸图像识别方法的实现流程示意图;
图4是本申请实施例四提供的模糊人脸图像识别方法的实现流程示意图;
图5是本申请实施例五提供的模糊人脸图像识别装置的结构示意图;
图6是本申请实施例六提供的终端设备的示意图。
本发明的实施方式
以下描述中,为了说明而不是为了限定,提出了诸如特定***结构、技术之类的具体细节,以便透彻理解本申请实施例。然而,本领域的技术人员应当清楚,在没有这些具体细节的其它实施例中也可以实现本申请。在其它情况中,省略对众所周知的***、装置、电路以及方法的详细说明,以免不必要的细节妨碍本申请的描述。
为了说明本申请所述的技术方案,下面通过具体实施例来进行说明。
为了便于理解本申请,此处先对本申请实施例进行简要说明,由于实际情况中造成人脸图像模糊的可能因素很多,从而导致实际应用中可能存在的模糊人脸图像类型也有多种,例如物体运动导致的运动模糊、光线不佳导致的光线模糊以及由于拍摄分辨率较低导致的分辨率模糊,现有技术中虽然也有一些模糊图像识别方法,但都是针对分辨率较低的模糊图像的识别,且都没有结合人脸本身进行设计,从而使得现有技术对不同类型模糊人脸图像识别的准确性难以得到有效地保障。
为了实现对不同类型模糊人脸图像的兼容识别以及提高识别准确率,本申请实施例中首先会预先采集各种不同类型的模糊人脸图像样本,并基于这些样本来进行模糊识别模型的训练,从而得到可以进行模糊图像识别的模糊识别模型,在此基础上,一方面通过预先对待处理图像进行人脸轮廓裁剪和图像尺寸调整,既保留了关键人脸特征,又使得对人脸图像模糊的判断能集中于人脸本身,减小了非人脸区域的清晰度干扰,另一方面,再通过基于训练好的模糊识别模型对待处理人脸区域图像进行模糊识别,实现了对多种人脸模糊类型的兼容性准确识别,准确判断待处理人脸区域是否模糊,同时为了防止人脸轮廓识别和模糊识别模型识别中可能存在的干扰导致识别误差增大,本申请实施例还进一步地对判断为模糊图像的待 处理图像进行是否包含人脸的二次校验,并在包含人脸概率较小的情况下,即各种模糊因素导致人脸识别准确率较低的情况下,才最终判定待处理图像为模糊人脸图像,从而极大地提升了最终对不同类型的模糊人脸图像识别的兼容性和准确率。详述如下:
图1示出了本申请实施例一提供的模糊人脸图像识别方法的实现流程图,详述如下:
S101,对待处理图像进行人脸轮廓裁剪并调整至预设图像尺寸,得到对应的待处理人脸区域图像。
实际情况中,即使是专门针对人脸进行拍摄得到的图像,例如证件照和自拍照等,其中也会包含大量非人脸区域,而传统模糊识别方法都是直接针对整张图像进行模糊识别,此时非人脸区域的模糊程度将极大地影响最终的模糊判断,例如在自拍过程中经常会使用到背景虚化技术,这使得最终得到的图像中非人脸的背景区域极为模糊,现有技术在进行模糊识别时极容易将整张图像识别为模糊图像。因此,为了提高对模糊人脸图像识别的准确性,同时保留关键人脸特征,本申请实施例首先会对待处理图像进行人脸轮廓裁剪,即识别出待处理图像中的人脸区域,并将其裁剪为独立的待处理人脸区域图像。同时,由于不同待处理图像中包含的人脸区域图像尺寸会存在一定差异,不便于后续模糊识别模型的处理识别,因此本申请实施例会统一将提取出的人脸区域进行尺寸缩放,调整至预设的图像尺寸,其中具体的图像尺寸可由技术人员根据实际需求进行设定,优选地,可以设置为112×112像素。
应当说明地,由于本申请实施例本身就是需要识别模糊的人脸图像,因此待处理图像中包含的人脸区域也极有可能是模糊的,从而使得对人脸区域的识别本身就难以精确化,若进行人脸识别的要求过高极有可能导致对一些模糊人脸区域的无法识别,从而无法正常对待处理图像进行人脸轮廓裁剪,因此本申请实施例在进行人脸轮廓裁剪时,应当采用一些精确度要求较低的人脸识别方法进行人脸识别查找并裁剪,包括但不限于如人脸轮廓定位算法来进行人脸轮廓查找,并将满足人脸轮廓的区域图像识别为人脸区域进行裁决,或者将一些传统的人脸识别算法中的人脸识别判断阈值减小,从而提高对模糊人脸的识别能力。
作为本申请的一个实施例,为了防止其他物体的干扰,使得人脸识别时得到多个人脸区域,如类似人脸轮廓的物体和拍摄时不小心拍到的非目标人脸的其他人脸等,本申请实施例在进行人脸区域识别时,仅会将其中满足人脸识别要求且像素点最多的区域识别为人脸区域,并进行后续的裁剪。
S102,将待处理人脸区域图像输入至预先训练好的模糊识别模型中进行处理,识别待处理人脸区域图像是否为模糊图像,其中,模糊识别模型为预先基于清晰人脸区域图像样本和多种模糊类型的模糊人脸区域图像样本进行训练得到的识别模型,用于识别图像是否为模糊图像。
在本申请实施例中,为了实现对不同类型的模糊人图像的兼容识别,会预先采集大量的各种类型的模糊人脸图像样本,并以之进行模型训练得到对应的模糊识别模型,其中,模糊识别模型的模型种类包括但不限于如二分类模型和神经网络模型,具体由技术人员根据实际需求选定,此处不予限定,同时具体的模型训练方法也可有技术人员自行设定,或者也可参考本申请实施例二及其他本申请相关实施例内容。
在预先训练好模糊识别模型的基础上,本申请实施例会将裁剪得到的待处理人脸区域图像输入至模糊识别模型中,利用模型进行处理识别待处理人脸区域是否为模糊图像。
S103,若待处理人脸区域图像为模糊图像,则基于预设的人脸识别模型对待处理图像进行人脸识别,判断待处理图像包含人脸的概率。
S104,若待处理图像包含人脸的概率小于预设人脸概率阈值,则判定待处理图像为模糊人脸图像。
由于是否模糊本身就是一个相对的概念(即模糊到何种程度才可以判定为模糊图像),即使在进行模糊识别模型训练时也是依赖于技术人员对图像样本标记的标签来进行类别划分,因此不同技术人员训练得到的模糊识别模型,在进行模糊识别判定的标准上可能会存在一定的差异,同时实际情况中,随着各种人脸处理算法的不断发展,对模糊程度相对不是很高的人脸图像的识别处理能力也得到了极大的提升,此时若将模糊程度较低的人脸图像也判定为模糊人脸图像,会使得对已采集好的人脸图像数据提出较高的更新需求,从而对实际应用人脸图像采集带来较大的干扰,使得采集的效率降低,因此,在实际应用中,对人脸图像是否模糊的判断需求应当更偏向于将模糊程度较高的人脸图像挑选出来。正是基于上述这些实际应用需求,本申请实施例在将待处理人脸区域图像识别为模糊图像时,并不会直接将其判定为模糊人脸图像,而是会对其进行二次校验,以保证最终识别结果能满足实际应用的需求,具体而言:考虑到无论何种模糊干扰因素最终都会导致人脸区域本身变得模糊,因此本申请实施例在二次检验时,会对判定为模糊图像的待处理人脸区域图像进行人脸识别得到其内包含人脸的概率,对于概率小于预设人脸概率阈值情况,即说明该待处理人脸区域图像中人脸模糊程度较高,难以正常识别出其中的人脸,因此本申请实施例此时会直接判定其内人脸模糊,即待处理图像为模糊人脸图像。其中,具体使用的人脸识别模型,此处不予限定,但与步骤S101不同的是,这里需要对待处理人脸区域图像进行相对较为精确的人脸查找匹配,以判断其模糊程度如何,因此需要选用一些精确度较高的人脸识别方法进行人脸识别,包括但不限于如一些基于深度学习模型训练得到的人脸识别模型,或者基于人脸特征点匹配的人脸识别模型。人脸概率阈值的具体值可由技术人员自行设定,或者参考本申请实施例四进行处理得到,此处不予限定。
本申请实施例中首先预先采集各种不同类型的模糊人脸图像样本,并基于这些样本来进行模糊识别模型的训练,从而得到可以进行模糊图像识别的模糊识别模型,在此基础上,一方面通过预先对待处理图像进行人脸轮廓裁剪和图像尺寸调整,既保留了关键人脸特征,又使得对人脸图像模糊的判断能集中于人脸本身,减小了非人脸区域的清晰度干扰,另一方面,再通过基于训练好的模糊识别模型对待处理人脸区域图像进行模糊识别,实现了对多种人脸模糊类型的兼容性准确识别,准确判断待处理人脸区域是否模糊,为了防止人脸轮廓识别和模糊识别模型识别中可能存在的干扰导致识别误差增大,同时满足实际应用的需求,仅将模糊程度较高的人脸图像挑选出来,本申请实施例还进一步地对判断为模糊图像的待处理图像进行是否包含人脸的二次校验,并在包含人脸概率较小的情况下,即人脸模糊程度较高难以正常识别出其中人脸的情况下,才最终判定待处理图像为模糊人脸图像,从而极大地提升了最终对不同类型的模糊人脸图像识别的兼容性和准确率。
作为对模糊识别模型进行训练构建的一种具体实现方式,如图2所示,在本申请实施例一之前,本申请实施例二包括:
S201,获取清晰人脸图像样本以及多种模糊类型的模糊人脸图像样本。
其中对人脸图像样本的获取方式,既可以是从网站爬取也可以是技术人员自行拍摄等,此处不予限定,但应当保证获取的模糊人脸图像样本中包含光线模糊类型、运动模糊类型和分辨率模糊类型的模糊人脸图像样本,以保证后续训练出的模型的有效性,同时,三种类型模糊人脸图像样本的数量应大致接近或相同。
作为本申请获取模糊人脸图像样本的一种具体实现方式,本申请实施例三中,由技术人员预先在不同的场景条件下对人脸进行图像采集并储存,以得到所需的不同类型的模糊人脸图像样本,如图3所示,包括:
S301,获取在不同光线条件下对人脸进行图像采集得到的第一预设数量光线模糊类型的模糊人脸图像样本。
其中,为了丰富不同模糊程度的模糊人脸图像样本,本申请实施例中可以调节不同的光线强度同时使用多种光线类型来改变采集的环境,并进行人脸图像采集。
S302,获取摄像头在与人体多种不同相对移动速度下拍摄的第二预设数量运动模糊类型的模糊人脸图像样本。
同理,本申请实施例中,也可以调节不同的相对移动速度来采集人脸图像,以得到运动模糊类型对应的不同模糊程度模糊人脸图像样本。
S303,获取清晰人脸图像样本,并对清晰人脸图像样本进行像素模糊化处理,得到第三预设数量分辨率模糊类型的模糊人脸图像样本。
这里的像素模糊处理方式包括但不限于如使用均值模糊、中值模糊、高斯模糊、双边模糊和缩放模糊对清晰人脸图像样本进行模糊处理,得到对应的分辨率模糊类型的模糊人脸图像样本。其中,为了保证样本数据之间的平衡,本申请实施例中三种类型模糊人脸图像样本的数量应大致接近或相同,具体而言,第一预设数量、第二预设数量和第三预设数量的值可由技术人员自行设定。
作为本申请的一个优选实施例,在本申请实施例三的基础上,本申请实施例还包括:爬取预设的网站,得到第四预设数量的模糊人脸图像样本。其中,第四预设数量、第一预设数量、第二预设数量和第三预设数量依次减小。
由于所需的样本数量较多,全部拍摄/处理难度较大且样本随机性不够,难以保证最终的效果,因此需要爬取实际生活中可能出现的模糊人脸图像,但实际情况中,可爬取到的模糊人脸数据里面包含的模糊类型是难以控制的,且根据对实际网站可爬取的图像数据进行分析后发现,各种类的模糊人脸图像数量分别情况为:分辨率模糊类型>运动模糊类型>光线模糊类型,因此在实际爬取的基础上,为了尽可能地平衡三种类型的模糊人脸图像数量,本申请实施例中中对第一预设数量、第二预设数量和第三预设数量的需求发生了变化,需要满足第四预设数量>第一预设数量>第二预设数量>第三预设数量,其中,具体的数量大小需由技术人员根据实际爬取情况设定。
S202,分别对清晰人脸图像样本和模糊人脸图像样本进行人脸轮廓裁剪并调整至预设图像尺寸,得到对应的清晰人脸区域图像样本和模糊人脸区域图像样本。
由于本申请实施例中需要采集大量的人脸图像样本并进行模型训练,为了减小模型训练的工作量和速度,本申请实施例中会对清晰人脸图像样本和模糊人脸图像样本均进行人脸轮廓裁剪,同时在本申请实施例中需要进行的是模糊人脸识别而不是精细的人脸识别或特征提取等,因此并不需要过多的人脸区域数据,因此在进行尺寸选取时本申请实施例优选选用较小的尺寸,以在保证模糊识别的基础上,尽可能地增加模型训练的速度以及本申请实施例一中对待处理图像的模糊人脸图像识别速度。其中,具体的人脸轮廓裁剪方法可参考本申请实施例一中相关说明,此处不予赘述。
S203,基于清晰人脸区域图像样本和模糊人脸区域图像样本对预设的二分类模型进行训练,并计算分类结果交叉熵的损失值。
作为本申请实施例二中计算分类结果交叉熵的损失值的一种具体实现方式,包括:
基于以下公式(1)计算分类结果交叉熵的损失值:
Figure PCTCN2019103159-appb-000001
其中,L为损失值,n为人脸区域图像样本的总数量,x i和y i分别是第i张人脸区域图像样本的实际标签值和分类标签值,m i为第i张人脸区域图像样本的中包含的人脸数量,人脸区域图像样本包括清晰人脸区域图像样本和模糊人脸区域图像样本。
S204,若损失值大于预设损失阈值,基于梯度下降法对二分类模型进行更新训练,直至损失值小于或等于预设损失阈值或迭代次数达到预设次数阈值,完成训练,得到模糊识别模型。
其中,具体采用的二分类模型此处不予限定,包括但不限于如基于逻辑回归模型的二分类模型和基于机器学习的二分类模型等,可由技术人员根据实际需求自行设置。
作为本申请实施例四,考虑到上述本申请实施例一至三在进行模糊图像的二次检验时,检验的准确性会直接受到所采用的人脸识别模型的准确性以及所采用的人脸概率阈值有效性影响,若由技术人员直接设定一个固定的人脸概率阈值进行判定,检验的准确性极易受到技术人员的主观性影响,从而使得准确性难以得到保障,因此,为了提高检验的有效性,如图4所示,包括:
S401,基于人脸识别模型对清晰人脸图像样本和模糊人脸图像样本进行人脸识别,得到每张清晰人脸图像样本和模糊人脸图像样本分别对应的包含人脸的概率。
S402,基于每张清晰人脸图像样本和模糊人脸图像样本分别对应的包含人脸的概率,确定预设人脸概率阈值。
考虑到实际应用中无论哪种人脸识别算法都不可能实现百分百准确的模糊识别,同时为了得到有效合理的人脸概率阈值,本申请实施例中会利用本申请实施例一实际选定的人脸识别模型,来对实际获取到的人脸图像样本进行识别,并将识别的结果与实际每张人脸图像样本的标签进行比对,选取出其中可以使得二分类效果最优的包含人脸概率作为本申请实施例一中的人脸概率阈值,例如,对每个人脸图像样本进行人脸识别得到对应的包含人脸概率后,发现当选取包含人脸概率=70%时,二分类的效果最优,此时本申请实施例就会将70%作为本申请实施例一中的人脸概率阈值。
本申请实施例中,通过实际使用的人脸识别模型和人脸图像样本进行人脸概率阈值的计算选取,可以尽可能地避免不同人脸识别模型识别准确性高低给本申请造成的识别不准确的情况,使得实际应用中无论技术人员选取什么样的人脸识别模型,本申请实施例都可以很好地根据人脸识别模型对实际样本的识别能力来确定出适宜的阈值,从而使得本申请实施例一对清晰人脸识别的能力实现自适应最优化。
对应于上文实施例的方法,图5示出了本申请实施例提供的模糊人脸图像识别装置的结构框图,为了便于说明,仅示出了与本申请实施例相关的部分。图5示例的模糊人脸图像识 别装置可以是前述实施例一提供的模糊人脸图像识别方法的执行主体。
参照图5,该模糊人脸图像识别装置包括:
人脸裁剪模块51,用于对待处理图像进行人脸轮廓裁剪并调整至预设图像尺寸,得到对应的待处理人脸区域图像。
模糊图像识别模块52,用于将所述待处理人脸区域图像输入至预先训练好的模糊识别模型中进行处理,识别所述待处理人脸区域图像是否为模糊图像,其中,模糊识别模型为预先基于清晰人脸区域图像样本和多种模糊类型的模糊人脸区域图像样本进行训练得到的识别模型,用于识别图像是否为模糊图像。
人脸识别模块53,用于若所述待处理人脸区域图像为模糊图像,则基于预设的人脸识别模型对所述待处理图像进行人脸识别,判断所述待处理图像包含人脸的概率。
模糊人脸识别模块54,用于若所述待处理图像包含人脸的概率小于预设人脸概率阈值,则判定所述待处理图像为模糊人脸图像。
进一步地,该模糊人脸图像识别装置,还包括:
样本获取模块,用于获取清晰人脸图像样本以及多种模糊类型的模糊人脸图像样本。
样本裁剪模块,用于分别对所述清晰人脸图像样本和所述模糊人脸图像样本进行人脸轮廓裁剪并调整至所述预设图像尺寸,得到对应的所述清晰人脸区域图像样本和所述模糊人脸区域图像样本。
模型训练模块,用于基于所述清晰人脸区域图像样本和所述模糊人脸区域图像样本对预设的二分类模型进行训练,并计算分类结果交叉熵的损失值。
迭代训练模块,用于若所述损失值大于预设损失阈值,基于梯度下降法对所述二分类模型进行更新训练,直至所述损失值小于或等于所述预设损失阈值或迭代次数达到预设次数阈值,完成训练,得到所述模糊识别模型。
进一步地,样本获取模块,包括:
获取在不同光线条件下对人脸进行图像采集得到的第一预设数量光线模糊类型的所述模糊人脸图像样本。
获取摄像头在与人体多种不同相对移动速度下拍摄的第二预设数量运动模糊类型的所述模糊人脸图像样本。
获取所述清晰人脸图像样本,并对所述清晰人脸图像样本进行像素模糊化处理,得到第三预设数量分辨率模糊类型的所述模糊人脸图像样本。
进一步地,样本获取模块,还包括:
爬取预设的网站,得到第四预设数量的所述模糊人脸图像样本。其中,第四预设数量、 第一预设数量、第二预设数量和第三预设数量依次减小。
进一步地,模型训练模块,包括
基于以下公式(1)计算所述分类结果交叉熵的损失值:
Figure PCTCN2019103159-appb-000002
其中,L为损失值,n为人脸区域图像样本的总数量,x i和y i分别是第i张人脸区域图像样本的实际标签值和分类标签值,m i为第i张人脸区域图像样本的中包含的人脸数量,人脸区域图像样本包括清晰人脸区域图像样本和模糊人脸区域图像样本。
进一步地,该模糊人脸图像识别装置,还包括:
基于所述人脸识别模型对所述清晰人脸图像样本和所述模糊人脸图像样本进行人脸识别,得到每张所述清晰人脸图像样本和所述模糊人脸图像样本分别对应的包含人脸的概率。
基于每张所述清晰人脸图像样本和所述模糊人脸图像样本分别对应的包含人脸的概率,确定所述预设人脸概率阈值。
本申请实施例提供的模糊人脸图像识别装置中各模块实现各自功能的过程,具体可参考前述图1所示实施例一的描述,此处不再赘述。
应理解,上述实施例中各步骤的序号的大小并不意味着执行顺序的先后,各过程的执行顺序应以其功能和内在逻辑确定,而不应对本申请实施例的实施过程构成任何限定。还应理解的是,虽然术语“第一”、“第二”等在文本中在一些本申请实施例中用来描述各种元素,但是这些元素不应该受到这些术语的限制。这些术语只是用来将一个元素与另一元素区分开。例如,第一表格可以被命名为第二表格,并且类似地,第二表格可以被命名为第一表格,而不背离各种所描述的实施例的范围。第一表格和第二表格都是表格,但是它们不是同一表格。
图6是本申请一实施例提供的终端设备的示意图。如图6所示,该实施例的终端设备6包括:处理器60、存储器61,所述存储器61中存储有可在所述处理器60上运行的计算机可读指令62。所述处理器60执行所述计算机可读指令62时实现上述各个模糊人脸图像识别方法实施例中的步骤,例如图1所示的步骤101至105。或者,所述处理器60执行所述计算机可读指令62时实现上述各装置实施例中各模块/单元的功能,例如图5所示模块51至54的功能。
所述终端设备6可以是桌上型计算机、笔记本、掌上电脑及云端服务器等计算设备。所述终端设备可包括,但不仅限于,处理器60、存储器61。本领域技术人员可以理解,图6仅仅是终端设备6的示例,并不构成对终端设备6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,例如所述终端设备还可以包括输入发送设备、网 络接入设备、总线等。
所称处理器60可以是中央处理单元(Central Processing Unit,CPU),还可以是其他通用处理器、数字信号处理器(Digital Signal Processor,DSP)、专用集成电路(Application Specific Integrated Circuit,ASIC)、现成可编程门阵列(Field-Programmable Gate Array,FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。
所述存储器61可以是所述终端设备6的内部存储单元,例如终端设备6的硬盘或内存。所述存储器61也可以是所述终端设备6的外部存储设备,例如所述终端设备6上配备的插接式硬盘,智能存储卡(Smart Media Card,SMC),安全数字(Secure Digital,SD)卡,闪存卡(Flash Card)等。进一步地,所述存储器61还可以既包括所述终端设备6的内部存储单元也包括外部存储设备。所述存储器61用于存储所述计算机可读指令以及所述终端设备所需的其他程序和数据。所述存储器61还可以用于暂时地存储已经发送或者将要发送的数据。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。
所述集成的模块/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,本申请实现上述实施例方法中的全部或部分流程,也可以通过计算机可读指令来指令相关的硬件来完成,所述的计算机可读指令可存储于一计算机可读存储介质中,该计算机可读指令在被处理器执行时,可实现上述各个方法实施例的步骤。
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,是可以通过计算机可读指令来指令相关的硬件来完成,所述的计算机可读指令可存储于一非易失性计算机可读取存储介质中,该计算机可读指令在执行时,可包括如上述各方法的实施例的流程。其中,本申请所提供的各实施例中所使用的对存储器、存储、数据库或其它介质的任何引用,均可包括非易失性和/或易失性存储器。非易失性存储器可包括只读存储器(ROM)、可编程ROM(PROM)、电可编程ROM(EPROM)、电可擦除可编程ROM(EEPROM)或闪存。易失性存储器可包括随机存取存储器(RAM)或者外部高速缓冲存储器。作为说明而非局限,RAM以多种形式可得,诸如静态RAM(SRAM)、动态RAM(DRAM)、同步DRAM(SDRAM)、双数据率SDRAM(DDRSDRAM)、增强型SDRAM(ESDRAM)、同步链路(Synchlink)DRAM(SLDRAM)、存储器总线(Rambus)直接RAM(RDRAM)、直接存储器总线动态RAM(DRDRAM)、以及存储器总线动态RAM(RDRAM)等。
以上所述实施例仅用以说明本申请的技术方案,而非对其限制;尽管参照前述实施例对本申请进行了详细的说明,本领域的普通技术人员应当理解:其依然可以对前述各实施例所记载的技术方案进行修改,或者对其中部分技术特征进行等同替换;而这些修改或者替换,并不使对应技术方案的本质脱离本申请各实施例技术方案的精神和范围,均应包含在本申请的保护范围之内。

Claims (20)

  1. 一种模糊人脸图像识别方法,其特征在于,包括:
    对待处理图像进行人脸轮廓裁剪并调整至预设图像尺寸,得到对应的待处理人脸区域图像;
    将所述待处理人脸区域图像输入至预先训练好的模糊识别模型中进行处理,识别所述待处理人脸区域图像是否为模糊图像,其中,模糊识别模型为预先基于清晰人脸区域图像样本和多种模糊类型的模糊人脸区域图像样本进行训练得到的识别模型,用于识别图像是否为模糊图像;
    若所述待处理人脸区域图像为模糊图像,则基于预设的人脸识别模型对所述待处理图像进行人脸识别,判断所述待处理图像包含人脸的概率;
    若所述待处理图像包含人脸的概率小于预设人脸概率阈值,则判定所述待处理图像为模糊人脸图像。
  2. 如权利要求1所述的模糊人脸图像识别方法,其特征在于,对所述模糊识别模型的训练包括:
    获取清晰人脸图像样本以及多种模糊类型的模糊人脸图像样本;
    分别对所述清晰人脸图像样本和所述模糊人脸图像样本进行人脸轮廓裁剪并调整至所述预设图像尺寸,得到对应的所述清晰人脸区域图像样本和所述模糊人脸区域图像样本;
    基于所述清晰人脸区域图像样本和所述模糊人脸区域图像样本对预设的二分类模型进行训练,并计算分类结果交叉熵的损失值;
    若所述损失值大于预设损失阈值,基于梯度下降法对所述二分类模型进行更新训练,直至所述损失值小于或等于所述预设损失阈值或迭代次数达到预设次数阈值,完成训练,得到所述模糊识别模型。
  3. 如权利要求2所述的模糊人脸图像识别方法,其特征在于,所述模糊类型包括运动模糊、光线模糊和分辨率模糊,所述获取清晰人脸图像样本以及多种模糊类型的模糊人脸图像样本,包括:
    获取在不同光线条件下对人脸进行图像采集得到的第一预设数量光线模糊类型的所述模糊人脸图像样本;
    获取摄像头在与人体多种不同相对移动速度下拍摄的第二预设数量运动模糊类型的所述模糊人脸图像样本;
    获取所述清晰人脸图像样本,并对所述清晰人脸图像样本进行像素模糊化处理,得到 第三预设数量分辨率模糊类型的所述模糊人脸图像样本。
  4. 如权利要求3所述的模糊人脸图像识别方法,其特征在于,所述模糊类型包括运动模糊、光线模糊和分辨率模糊,所述获取清晰人脸图像样本以及多种模糊类型的模糊人脸图像样本,还包括:
    爬取预设的网站,得到第四预设数量的所述模糊人脸图像样本;其中,第四预设数量、第一预设数量、第二预设数量和第三预设数量依次减小。
  5. 如权利要求2所述的模糊人脸图像识别方法,其特征在于,所述计算分类结果交叉熵的损失值,包括
    基于以下公式计算所述分类结果交叉熵的损失值:
    Figure PCTCN2019103159-appb-100001
    其中,L为损失值,n为人脸区域图像样本的总数量,x i和y i分别是第i张人脸区域图像样本的实际标签值和分类标签值,m i为第i张人脸区域图像样本的中包含的人脸数量,人脸区域图像样本包括清晰人脸区域图像样本和模糊人脸区域图像样本。
  6. 如权利要求2至5任意一项所述的模糊人脸图像识别方法,其特征在于,在对待处理图像进行人脸轮廓裁剪并调整至预设图像尺寸之前,还包括:
    基于所述人脸识别模型对所述清晰人脸图像样本和所述模糊人脸图像样本进行人脸识别,得到每张所述清晰人脸图像样本和所述模糊人脸图像样本分别对应的包含人脸的概率;
    基于每张所述清晰人脸图像样本和所述模糊人脸图像样本分别对应的包含人脸的概率,确定所述预设人脸概率阈值。
  7. 一种模糊人脸图像识别装置,其特征在于,包括:
    人脸裁剪模块,用于对待处理图像进行人脸轮廓裁剪并调整至预设图像尺寸,得到对应的待处理人脸区域图像;
    模糊图像识别模块,用于将所述待处理人脸区域图像输入至预先训练好的模糊识别模型中进行处理,识别所述待处理人脸区域图像是否为模糊图像,其中,模糊识别模型为预先基于清晰人脸区域图像样本和多种模糊类型的模糊人脸区域图像样本进行训练得到的识别模型,用于识别图像是否为模糊图像;
    人脸识别模块,用于若所述待处理人脸区域图像为模糊图像,则基于预设的人脸识别模型对所述待处理图像进行人脸识别,判断所述待处理图像包含人脸的概率;
    模糊人脸识别模块,用于若所述待处理图像包含人脸的概率小于预设人脸概率阈值,则判定所述待处理图像为模糊人脸图像。
  8. 如权利要求7所述的模糊人脸图像识别装置,其特征在于,还包括:
    样本获取模块,用于获取清晰人脸图像样本以及多种模糊类型的模糊人脸图像样本;
    样本裁剪模块,用于分别对所述清晰人脸图像样本和所述模糊人脸图像样本进行人脸轮廓裁剪并调整至所述预设图像尺寸,得到对应的所述清晰人脸区域图像样本和所述模糊人脸区域图像样本;
    模型训练模块,用于基于所述清晰人脸区域图像样本和所述模糊人脸区域图像样本对预设的二分类模型进行训练,并计算分类结果交叉熵的损失值;
    迭代训练模块,用于若所述损失值大于预设损失阈值,基于梯度下降法对所述二分类模型进行更新训练,直至所述损失值小于或等于所述预设损失阈值或迭代次数达到预设次数阈值,完成训练,得到所述模糊识别模型。
  9. 如权利要求8所述的模糊人脸图像识别装置,其特征在于,样本获取模块,包括:
    获取在不同光线条件下对人脸进行图像采集得到的第一预设数量光线模糊类型的所述模糊人脸图像样本。
    获取摄像头在与人体多种不同相对移动速度下拍摄的第二预设数量运动模糊类型的所述模糊人脸图像样本。
    获取所述清晰人脸图像样本,并对所述清晰人脸图像样本进行像素模糊化处理,得到第三预设数量分辨率模糊类型的所述模糊人脸图像样本。
  10. 如权利要求9所述的模糊人脸图像识别装置,其特征在于,样本获取模块,还包括:
    爬取预设的网站,得到第四预设数量的所述模糊人脸图像样本。其中,第四预设数量、第一预设数量、第二预设数量和第三预设数量依次减小。
  11. 如权利要求8所述的模糊人脸图像识别装置,其特征在于,,模型训练模块,包括基于以下公式计算所述分类结果交叉熵的损失值:
    Figure PCTCN2019103159-appb-100002
    其中,L为损失值,n为人脸区域图像样本的总数量,x i和y i分别是第i张人脸区域图像样本的实际标签值和分类标签值,m i为第i张人脸区域图像样本的中包含的人脸数量,人脸区域图像样本包括清晰人脸区域图像样本和模糊人脸区域图像样本。
  12. 如权利要求7至11任意一项所述的模糊人脸图像识别装置,其特征在于,还包括:
    基于所述人脸识别模型对所述清晰人脸图像样本和所述模糊人脸图像样本进行人脸识别,得到每张所述清晰人脸图像样本和所述模糊人脸图像样本分别对应的包含人脸的概率。
    基于每张所述清晰人脸图像样本和所述模糊人脸图像样本分别对应的包含人脸的概率, 确定所述预设人脸概率阈值。
  13. 一种终端设备,其特征在于,所述终端设备包括存储器、处理器,所述存储器上存储有可在所述处理器上运行的计算机可读指令,所述处理器执行所述计算机可读指令时实现如下步骤:
    对待处理图像进行人脸轮廓裁剪并调整至预设图像尺寸,得到对应的待处理人脸区域图像;
    将所述待处理人脸区域图像输入至预先训练好的模糊识别模型中进行处理,识别所述待处理人脸区域图像是否为模糊图像,其中,模糊识别模型为预先基于清晰人脸区域图像样本和多种模糊类型的模糊人脸区域图像样本进行训练得到的识别模型,用于识别图像是否为模糊图像;
    若所述待处理人脸区域图像为模糊图像,则基于预设的人脸识别模型对所述待处理图像进行人脸识别,判断所述待处理图像包含人脸的概率;
    若所述待处理图像包含人脸的概率小于预设人脸概率阈值,则判定所述待处理图像为模糊人脸图像。
  14. 如权利要求13所述的终端设备,其特征在于,对所述模糊识别模型的训练包括:
    获取清晰人脸图像样本以及多种模糊类型的模糊人脸图像样本;
    分别对所述清晰人脸图像样本和所述模糊人脸图像样本进行人脸轮廓裁剪并调整至所述预设图像尺寸,得到对应的所述清晰人脸区域图像样本和所述模糊人脸区域图像样本;
    基于所述清晰人脸区域图像样本和所述模糊人脸区域图像样本对预设的二分类模型进行训练,并计算分类结果交叉熵的损失值;
    若所述损失值大于预设损失阈值,基于梯度下降法对所述二分类模型进行更新训练,直至所述损失值小于或等于所述预设损失阈值或迭代次数达到预设次数阈值,完成训练,得到所述模糊识别模型。
  15. 如权利要求14所述的终端设备,其特征在于,所述模糊类型包括运动模糊、光线模糊和分辨率模糊,所述获取清晰人脸图像样本以及多种模糊类型的模糊人脸图像样本,包括:
    获取在不同光线条件下对人脸进行图像采集得到的第一预设数量光线模糊类型的所述模糊人脸图像样本;
    获取摄像头在与人体多种不同相对移动速度下拍摄的第二预设数量运动模糊类型的所述模糊人脸图像样本;
    获取所述清晰人脸图像样本,并对所述清晰人脸图像样本进行像素模糊化处理,得到 第三预设数量分辨率模糊类型的所述模糊人脸图像样本。
  16. 如权利要求15所述的终端设备,其特征在于,所述模糊类型包括运动模糊、光线模糊和分辨率模糊,所述获取清晰人脸图像样本以及多种模糊类型的模糊人脸图像样本,还包括:
    爬取预设的网站,得到第四预设数量的所述模糊人脸图像样本;其中,第四预设数量、第一预设数量、第二预设数量和第三预设数量依次减小。
  17. 如权利要求14所述的终端设备,其特征在于,所述计算分类结果交叉熵的损失值,包括
    基于以下公式计算所述分类结果交叉熵的损失值:
    Figure PCTCN2019103159-appb-100003
    其中,L为损失值,n为人脸区域图像样本的总数量,x i和y i分别是第i张人脸区域图像样本的实际标签值和分类标签值,m i为第i张人脸区域图像样本的中包含的人脸数量,人脸区域图像样本包括清晰人脸区域图像样本和模糊人脸区域图像样本。
  18. 一种计算机可读存储介质,所述计算机可读存储介质存储有计算机可读指令,其特征在于,所述计算机可读指令被至少一个处理器执行时实现如下步骤:
    对待处理图像进行人脸轮廓裁剪并调整至预设图像尺寸,得到对应的待处理人脸区域图像;
    将所述待处理人脸区域图像输入至预先训练好的模糊识别模型中进行处理,识别所述待处理人脸区域图像是否为模糊图像,其中,模糊识别模型为预先基于清晰人脸区域图像样本和多种模糊类型的模糊人脸区域图像样本进行训练得到的识别模型,用于识别图像是否为模糊图像;
    若所述待处理人脸区域图像为模糊图像,则基于预设的人脸识别模型对所述待处理图像进行人脸识别,判断所述待处理图像包含人脸的概率;
    若所述待处理图像包含人脸的概率小于预设人脸概率阈值,则判定所述待处理图像为模糊人脸图像。
  19. 根据权利要求18所述的计算机可读存储介质,其特征在于,对所述模糊识别模型的训练包括:
    获取清晰人脸图像样本以及多种模糊类型的模糊人脸图像样本;
    分别对所述清晰人脸图像样本和所述模糊人脸图像样本进行人脸轮廓裁剪并调整至所述预设图像尺寸,得到对应的所述清晰人脸区域图像样本和所述模糊人脸区域图像样本;
    基于所述清晰人脸区域图像样本和所述模糊人脸区域图像样本对预设的二分类模型进行训练,并计算分类结果交叉熵的损失值;
    若所述损失值大于预设损失阈值,基于梯度下降法对所述二分类模型进行更新训练,直至所述损失值小于或等于所述预设损失阈值或迭代次数达到预设次数阈值,完成训练,得到所述模糊识别模型。
  20. 根据权利要求19任意一项所述的计算机可读存储介质,其特征在于,所述模糊类型包括运动模糊、光线模糊和分辨率模糊,所述获取清晰人脸图像样本以及多种模糊类型的模糊人脸图像样本,包括:
    获取在不同光线条件下对人脸进行图像采集得到的第一预设数量光线模糊类型的所述模糊人脸图像样本;
    获取摄像头在与人体多种不同相对移动速度下拍摄的第二预设数量运动模糊类型的所述模糊人脸图像样本;
    获取所述清晰人脸图像样本,并对所述清晰人脸图像样本进行像素模糊化处理,得到第三预设数量分辨率模糊类型的所述模糊人脸图像样本。
PCT/CN2019/103159 2019-06-20 2019-08-29 一种模糊人脸图像识别方法、装置、终端设备及介质 WO2020252917A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910536758.X 2019-06-20
CN201910536758.XA CN110378235B (zh) 2019-06-20 2019-06-20 一种模糊人脸图像识别方法、装置及终端设备

Publications (1)

Publication Number Publication Date
WO2020252917A1 true WO2020252917A1 (zh) 2020-12-24

Family

ID=68250416

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/103159 WO2020252917A1 (zh) 2019-06-20 2019-08-29 一种模糊人脸图像识别方法、装置、终端设备及介质

Country Status (2)

Country Link
CN (1) CN110378235B (zh)
WO (1) WO2020252917A1 (zh)

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112766235A (zh) * 2021-02-24 2021-05-07 北京嘀嘀无限科技发展有限公司 人脸识别方法、装置、设备、存储介质和计算机程序产品
CN112949423A (zh) * 2021-02-07 2021-06-11 深圳市优必选科技股份有限公司 物体识别方法、物体识别装置及机器人
CN113436137A (zh) * 2021-03-12 2021-09-24 北京世纪好未来教育科技有限公司 一种图像清晰度识别方法、装置、设备及介质
CN113657178A (zh) * 2021-07-22 2021-11-16 浙江大华技术股份有限公司 人脸识别方法、电子设备和计算机可读存储介质
CN113705461A (zh) * 2021-08-30 2021-11-26 平安银行股份有限公司 人脸清晰度检测方法、装置、设备及存储介质
CN114549346A (zh) * 2022-01-27 2022-05-27 阿丘机器人科技(苏州)有限公司 模糊图像识别方法、装置、设备及存储介质
CN115171197A (zh) * 2022-09-01 2022-10-11 广州市森锐科技股份有限公司 一种高精度图像信息识别方法、***、设备及存储介质
CN117240607A (zh) * 2023-11-10 2023-12-15 北京云尚汇信息技术有限责任公司 一种基于安全计算机的安全认证方法

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110992395B (zh) * 2019-11-01 2023-08-18 北京达佳互联信息技术有限公司 图像训练样本的生成方法及装置、运动跟踪方法及装置
CN111222411B (zh) * 2019-11-28 2023-09-01 中国船舶重工集团公司第七一三研究所 一种激光发射安全快速报警方法和装置
CN111460939A (zh) * 2020-03-20 2020-07-28 深圳市优必选科技股份有限公司 一种去模糊的人脸识别方法、***和一种巡检机器人
CN111814702A (zh) * 2020-07-13 2020-10-23 安徽兰臣信息科技有限公司 一种基于成年人脸和儿童照特征空间映射关系的儿童人脸识别方法
CN111914939B (zh) * 2020-08-06 2023-07-28 平安科技(深圳)有限公司 识别模糊图像的方法、装置、设备及计算机可读存储介质
CN112001940B (zh) * 2020-08-21 2023-04-07 Oppo(重庆)智能科技有限公司 图像处理方法及装置、终端及可读存储介质
CN112395965A (zh) * 2020-11-06 2021-02-23 山东鲁能软件技术有限公司 一种基于电力内网的移动终端人脸识别***及方法
CN112669207A (zh) * 2020-12-21 2021-04-16 四川长虹电器股份有限公司 一种基于电视摄像头的人脸图像分辨率增强方法
CN112965604A (zh) * 2021-03-29 2021-06-15 深圳市优必选科技股份有限公司 手势识别方法、装置、终端设备及计算机可读存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106951825A (zh) * 2017-02-13 2017-07-14 北京飞搜科技有限公司 一种人脸图像质量评估***以及实现方法
CN107133948A (zh) * 2017-05-09 2017-09-05 电子科技大学 基于多任务卷积神经网络的图像模糊与噪声评测方法
EP3333768A1 (en) * 2016-12-07 2018-06-13 Samsung Electronics Co., Ltd. Method and apparatus for detecting target
CN108241821A (zh) * 2016-12-23 2018-07-03 北京三星通信技术研究有限公司 图像处理设备和方法
CN108288027A (zh) * 2017-12-28 2018-07-17 新智数字科技有限公司 一种图像质量的检测方法、装置及设备

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109829396B (zh) * 2019-01-16 2020-11-13 广州杰赛科技股份有限公司 人脸识别运动模糊处理方法、装置、设备及存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP3333768A1 (en) * 2016-12-07 2018-06-13 Samsung Electronics Co., Ltd. Method and apparatus for detecting target
CN108241821A (zh) * 2016-12-23 2018-07-03 北京三星通信技术研究有限公司 图像处理设备和方法
CN106951825A (zh) * 2017-02-13 2017-07-14 北京飞搜科技有限公司 一种人脸图像质量评估***以及实现方法
CN107133948A (zh) * 2017-05-09 2017-09-05 电子科技大学 基于多任务卷积神经网络的图像模糊与噪声评测方法
CN108288027A (zh) * 2017-12-28 2018-07-17 新智数字科技有限公司 一种图像质量的检测方法、装置及设备

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112949423A (zh) * 2021-02-07 2021-06-11 深圳市优必选科技股份有限公司 物体识别方法、物体识别装置及机器人
CN112949423B (zh) * 2021-02-07 2024-05-24 深圳市优必选科技股份有限公司 物体识别方法、物体识别装置及机器人
CN112766235A (zh) * 2021-02-24 2021-05-07 北京嘀嘀无限科技发展有限公司 人脸识别方法、装置、设备、存储介质和计算机程序产品
CN113436137A (zh) * 2021-03-12 2021-09-24 北京世纪好未来教育科技有限公司 一种图像清晰度识别方法、装置、设备及介质
CN113657178A (zh) * 2021-07-22 2021-11-16 浙江大华技术股份有限公司 人脸识别方法、电子设备和计算机可读存储介质
CN113705461A (zh) * 2021-08-30 2021-11-26 平安银行股份有限公司 人脸清晰度检测方法、装置、设备及存储介质
CN114549346A (zh) * 2022-01-27 2022-05-27 阿丘机器人科技(苏州)有限公司 模糊图像识别方法、装置、设备及存储介质
CN115171197A (zh) * 2022-09-01 2022-10-11 广州市森锐科技股份有限公司 一种高精度图像信息识别方法、***、设备及存储介质
CN117240607A (zh) * 2023-11-10 2023-12-15 北京云尚汇信息技术有限责任公司 一种基于安全计算机的安全认证方法
CN117240607B (zh) * 2023-11-10 2024-02-13 北京云尚汇信息技术有限责任公司 一种基于安全计算机的安全认证方法

Also Published As

Publication number Publication date
CN110378235B (zh) 2024-05-28
CN110378235A (zh) 2019-10-25

Similar Documents

Publication Publication Date Title
WO2020252917A1 (zh) 一种模糊人脸图像识别方法、装置、终端设备及介质
CN109492643B (zh) 基于ocr的证件识别方法、装置、计算机设备及存储介质
CN110826519B (zh) 人脸遮挡检测方法、装置、计算机设备及存储介质
CN109165589B (zh) 基于深度学习的车辆重识别方法和装置
CN110163114B (zh) 一种人脸角度及人脸模糊度分析方法、***和计算机设备
WO2022033150A1 (zh) 图像识别方法、装置、电子设备及存储介质
WO2019233297A1 (zh) 数据集的构建方法、移动终端、可读存储介质
WO2019232866A1 (zh) 人眼模型训练方法、人眼识别方法、装置、设备及介质
WO2018028255A1 (zh) 基于对抗网络的图像显著性检测方法
WO2020238054A1 (zh) Pdf文档中图表的定位方法、装置及计算机设备
WO2017096753A1 (zh) 人脸关键点跟踪方法、终端和非易失性计算机可读存储介质
CN111445459B (zh) 一种基于深度孪生网络的图像缺陷检测方法及***
CN111191568B (zh) 翻拍图像识别方法、装置、设备及介质
WO2023130648A1 (zh) 一种图像数据增强方法、装置、计算机设备和存储介质
WO2021143865A1 (zh) 定位方法及装置、电子设备、计算机可读存储介质
WO2019228040A1 (zh) 一种面部图像评分方法及摄像机
CN112926654A (zh) 预标注模型训练、证件预标注方法、装置、设备及介质
WO2021238586A1 (zh) 一种训练方法、装置、设备以及计算机可读存储介质
CN105335760A (zh) 一种图像数字字符识别方法
WO2020172870A1 (zh) 一种目标对象的移动轨迹确定方法和装置
CN111340022A (zh) 身份证信息识别方法、装置、计算机设备及存储介质
WO2021169642A1 (zh) 基于视频的眼球转向确定方法与***
US11893773B2 (en) Finger vein comparison method, computer equipment, and storage medium
WO2022134354A1 (zh) 车损检测模型训练、车损检测方法、装置、设备及介质
CN113706481A (zh) ***质量检测方法、装置、计算机设备和存储介质

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: 19934204

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

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 01.06.2022)

122 Ep: pct application non-entry in european phase

Ref document number: 19934204

Country of ref document: EP

Kind code of ref document: A1