CN114639156B - Depression angle face recognition method and system based on axial attention weight distribution network - Google Patents
Depression angle face recognition method and system based on axial attention weight distribution network Download PDFInfo
- Publication number
- CN114639156B CN114639156B CN202210534034.3A CN202210534034A CN114639156B CN 114639156 B CN114639156 B CN 114639156B CN 202210534034 A CN202210534034 A CN 202210534034A CN 114639156 B CN114639156 B CN 114639156B
- Authority
- CN
- China
- Prior art keywords
- attention
- weight
- axial
- feature
- features
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 32
- 239000013598 vector Substances 0.000 claims abstract description 39
- 238000000605 extraction Methods 0.000 claims abstract description 25
- 239000011159 matrix material Substances 0.000 claims abstract description 17
- 230000002776 aggregation Effects 0.000 claims abstract description 15
- 238000004220 aggregation Methods 0.000 claims abstract description 15
- 238000013527 convolutional neural network Methods 0.000 claims description 12
- 238000006243 chemical reaction Methods 0.000 claims description 10
- 238000012544 monitoring process Methods 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 5
- 230000004927 fusion Effects 0.000 claims description 4
- 230000001815 facial effect Effects 0.000 claims description 3
- 238000006116 polymerization reaction Methods 0.000 claims description 3
- 230000009286 beneficial effect Effects 0.000 abstract description 3
- 238000010586 diagram Methods 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 14
- 238000001514 detection method Methods 0.000 description 6
- 210000000887 face Anatomy 0.000 description 4
- 238000013459 approach Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 210000001061 forehead Anatomy 0.000 description 2
- 230000004075 alteration Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000007500 overflow downdraw method Methods 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 239000000047 product Substances 0.000 description 1
- 210000000697 sensory organ Anatomy 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a depression angle face recognition method and a depression angle face recognition system based on an axial attention weight distribution network, wherein firstly, CNN is utilized to carry out feature extraction on face image sequences with different angles and resolutions so as to generate a group of normalized feature vectors; secondly, self-adaptively distributing weights for the image characteristic sequences by utilizing a cascaded axial attention weight distribution module to obtain a weight vector (transverse) of each characteristic diagram and a weight vector (longitudinal) between the characteristics to form an axial weight matrix; and finally, carrying out weighted aggregation by using the axial weight to obtain a feature vector with stronger discriminability for identification. The axial matrix weight of the invention can represent the weight distribution result among the face sequence images with more fineness, so the fused features have stronger discriminability and are beneficial to more accurate face recognition at a depression angle.
Description
Technical Field
The invention belongs to the technical field of artificial intelligence, relates to a multi-view face recognition method of a monitoring video, and particularly relates to a depression face recognition method based on an axial attention weight distribution network.
Technical Field
With the increasing popularization of the application of the face recognition technology, various and complex application scenes put higher requirements on the multi-pose face detection and recognition technology. In a public video surveillance condition, a surveillance camera is generally installed at a high position to overlook a target, and a face image taken is a face with an angle of depression. Although face detection and recognition methods based on deep learning models are becoming more and more mature, it is still quite difficult to realize high-precision face detection and recognition in a top view.
Since most surveillance cameras are typically mounted at a height of about 3.5 meters, when a pedestrian approaches the surveillance camera, only the forehead of the top view can be seen, and complete information of five sense organs cannot be obtained. The process of face recognition is from detection, cutting, aligning to the process of extracting features and recognizing, if the obtained face pictures are all top views from top to bottom, only the forehead and partial mandibular angle contours can be seen, important facial features such as mouth, mandibular angle and the like are shielded, and integral facial information is lacked. Many existing recognition networks only have the capability of recognizing front faces or side faces, which causes that the networks cannot be completed in the first step of face recognition detection, and accurate detection, normalization alignment and cutting of the faces in images are difficult, so that corresponding face features cannot be extracted. The angle of depression face is different from the side face, does not have face symmetry, can not utilize information of face symmetry to supplement face features, can not utilize priori knowledge to provide missing features for a network, and therefore compared with common side face recognition, the angle of depression face recognition from top to bottom has higher technical difficulty.
In the process that the pedestrian approaches the monitoring camera from far to near, the shot far and near sight distance face images have different complementary characteristics. When the pedestrian is far away from the camera, the face image close to the front can be acquired, the face image has a complete face contour, and the spatial resolution of the image is insufficient due to the long distance. When a pedestrian approaches the camera, a high-resolution image rich in fine texture details can be acquired, but because the pedestrian is too close to the camera, only a face image with a large overlooking angle can be acquired.
Therefore, the invention provides a depression angle face recognition method based on an axial attention weight distribution network, which reasonably utilizes the characteristics of a high-resolution depression angle face image at a near position and a low-resolution front face image at a far position to carry out effective face recognition.
Disclosure of Invention
In order to solve the technical problems, the invention provides a face recognition method based on an axial weight distribution network by combining the thought of deep learning attention and the thought of multi-frame face recognition, which utilizes a self-adaptive axial weight distribution network to complete the extraction and fusion of the features of the face with different resolutions and different angles of depression, and self-adaptively distributes weights to different feature vectors of the same person; the weight matrix acquired by the axial weight distribution module can provide finer distribution information for the network, and is beneficial to fusing the front information of the far low-resolution front face image and the detail texture information of the near high-resolution face, so that the face features with stronger discriminability are obtained, and the recognition effect is improved.
The method adopts the technical scheme that: a depression angle face recognition method based on an axial attention weight distribution network comprises the following steps:
step 1: carrying out feature extraction on a depression angle face image to be recognized;
collecting a plurality of face images with different angles and different resolutions, and performing feature extraction on the face image sequences with different angles and different resolutions to generate a group of normalized feature vectors which respectively represent the features of the face images with different resolutions and different angles which may appear in a section of monitoring video;
the method comprises the following steps of (1) extracting features of a face image by adopting a CNN (convolutional neural network); the CNN network adopts an ArcFace backbone network and consists of ResNet 50; after the image passes through a feature extraction module, generating a feature vector with dimensions of 1 multiplied by 512 after standardization; thus, extract outnFeatures of the input image are obtainednA feature vectorF 1 ~F n ;
And 2, step: inputting the depression angle face image after feature extraction into a cascade axial attention weight distribution module for feature aggregation;
the cascade axial attention weight distribution module uses a cascade axial attention block with nonlinear transfer inside and is responsible for obtaining a weight matrix; the cascade axial attention block comprises a transverse attention block and a longitudinal attention block, wherein the transverse attention block consists of two kernel convolutions and a nonlinear conversion function, the convolution kernels are realized through an FC layer, the nonlinear conversion is realized through a Sigmoid activation function, a self-attention weight parameter of the feature is obtained, and the weight is distributed to the feature; the vertical attention block is composed of two Conv layers and a Softmax function and generates weights of weight relations among vertical features; multiplying the two axial weights to obtain an attention parameter in a matrix form, obtaining more exquisite attention information, distributing attention to the multi-frame image, and obtaining a final aggregation vector;
and 3, step 3: and calculating similarity by using the aggregated features through a cosine similarity function, and performing feature matching on different faces to realize face recognition.
The technical scheme adopted by the system of the invention is as follows: a depression angle face recognition system based on an axial attention weight distribution network comprises the following modules:
the module 1 is used for extracting the features of a depression angle human face image to be recognized;
collecting a plurality of face images with different angles and different resolutions, and performing feature extraction on the face image sequences with different angles and different resolutions to generate a group of normalized feature vectors which respectively represent the features of the face images with different resolutions and different angles which may appear in a section of monitoring video;
the method comprises the following steps of (1) extracting features of a face image by adopting a CNN network; the CNN network adopts an ArcFace backbone network and consists of ResNet 50; after the image passes through a feature extraction module, generating a feature vector with dimensions of 1 multiplied by 512 after standardization; thus, extract outnFeatures of the input image are obtainednA feature vectorF 1 ~F n ;
The module 2 is used for inputting the depression angle face image after the characteristic extraction into the cascade axial attention weight distribution module for characteristic aggregation;
the cascade axial attention weight distribution module uses a cascade axial attention block with nonlinear transfer inside and is responsible for obtaining a weight matrix; the cascade axial attention block comprises a transverse attention block and a longitudinal attention block, wherein the transverse attention block consists of two kernel convolutions and a nonlinear conversion function, the convolution kernels are realized through an FC layer, the nonlinear conversion is realized through a Sigmoid activation function, a self-attention weight parameter of the feature is obtained, and the weight is distributed to the feature; the vertical attention block is composed of two Conv layers and a Softmax function and generates weights of weight relations among vertical features; multiplying the two axial weights to obtain an attention parameter in a matrix form, obtaining more exquisite attention information, distributing attention to the multi-frame image, and obtaining a final polymerization vector;
and a module 3: and calculating similarity through a cosine similarity function by using the aggregated features, and performing feature matching on different human faces to realize human face recognition.
Compared with the existing face recognition method, the method has the following advantages and positive effects:
(1) the invention provides a method for reasonably utilizing face images with different qualities by utilizing a multi-frame fusion method aiming at the problem of face recognition caused by factors such as resolution ratio, angle change and the like in the existing real monitoring scene, and realizes more accurate face recognition.
(2) The cascade axial attention weight distribution module provided by the invention can be used for carrying out self-attention extraction and inter-feature attention extraction on the features, and the obtained attention weight is expressed in a matrix form, so that the image weight distribution result can be expressed more finely, and the features with stronger discriminability can be obtained.
Drawings
FIG. 1: a flow chart of an embodiment of the invention;
FIG. 2: the invention discloses a structure diagram of a cascade axial attention weight distribution module.
Detailed Description
The invention will be described in further detail with reference to the drawings and examples, so that those skilled in the art can understand and practice the invention, and it is to be understood that the examples are merely illustrative and explanatory of the invention and are not restrictive thereof.
Referring to fig. 1, the present invention provides a depression angle face recognition method based on an axial attention weight distribution network, including the following steps:
step 1: carrying out feature extraction on a depression angle face image to be recognized;
collecting a plurality of face images with different angles and different resolutions, and performing feature extraction on the face image sequences with different angles and different resolutions to generate a group of normalized feature vectors which respectively represent the features of the face images with different resolutions and different angles which may appear in a section of monitoring video;
in the embodiment, a CNN network is adopted to extract features of a face image; the CNN network adopts an ArcFace backbone network and consists of ResNet 50; after the image passes through a feature extraction module, generating a feature vector with dimensions of 1 multiplied by 512 after standardization; thus, extract outnFeatures of the input image are obtainednA feature vectorF 1 ~F n ;
Step 2: inputting the depression angle face image after feature extraction into a cascade axial attention weight distribution module for feature aggregation;
please refer to fig. 2, in the cascade axial attention weight distribution module of the present embodiment, a cascade axial attention block with nonlinear transfer is used inside, and is responsible for obtaining a weight matrix; the cascade axial attention block comprises a transverse attention block and a longitudinal attention block, wherein the transverse attention block consists of two kernel convolutions and a nonlinear conversion function, the convolution kernels are realized through an FC layer, the nonlinear conversion is realized through a Sigmoid activation function, a self-attention weight parameter of the feature is obtained, and the weight is distributed to the feature; the vertical attention block is composed of two Conv layers and a Softmax function and generates weights of weight relations among vertical features; multiplying the two axial weights to obtain the attention parameter in a matrix form, obtaining more exquisite attention information, distributing attention to the multi-frame image, and obtaining a final aggregation vector.
The feature aggregation process of this example includes the following steps:
step 2.1: inputting the features into a cascade axial attention weight distribution module to perform weight self-adaptive distribution, wherein the weight self-adaptive distribution comprises a weight vector (transverse) of each feature map and a weight vector (longitudinal) between the features to form an axial weight matrix;
inputting the extracted characteristics of the low-resolution face image and the high-resolution depression angle face image with different qualities into a cascade axial attention weight distribution module, and respectively extracting transverse and longitudinal attention parameters;
step 2.1.1: extracting a transverse attention parameter through a transverse attention weight distribution module, wherein the transverse attention weight distribution module comprises two FC layers and a Sigmoid function, and obtaining a self-attention weight A of each featureiWherein A isiSize and FiConsistently, the weight assignment results inside each feature are represented:
f is n feature vectors obtained by extracting features of the n input images in the step 1;
step 2.1.2: performing self-attention fusion on each feature by using the self-attention weight Ai to obtain a transverse weight distribution result S:
step 2.1.3: longitudinal weight distribution is carried out through a longitudinal attention weight distribution module, the weight among all the characteristics is distributed in the longitudinal process, the longitudinal attention extraction module comprises two Conv layers and a Softmax function, and the longitudinal attention distribution module obtains the weight distributionnWeight W between features representing different picture featuresiW represents the vertical weight assignment:
step 2.2: and carrying out weighted aggregation by using the axial weight to obtain the final aggregation characteristic vector with stronger discriminability.
In this embodiment, the features are aggregated using the obtained weight assignment results, and the aggregated features r are represented as follows:
the symbol "", indicates a dot product operation.
And step 3: and calculating similarity through a cosine similarity function by using the aggregated features, and performing feature matching on different human faces to realize human face recognition.
The invention combines the thought of deep learning attention and the thought of multi-frame face recognition, and provides a face recognition method based on an axial weight distribution network, which utilizes a self-adaptive axial weight distribution network to complete the extraction and fusion of the features of the face with different resolutions and different angles, and self-adaptively distributes weights to different feature vectors of the same person; the weight matrix obtained by the axial weight distribution module can provide more exquisite distribution information for a network, and is beneficial to fusing the front information of a far low-resolution front face image and the detail texture information of a near high-resolution face, so that more discriminative face features are obtained, and the recognition effect is improved.
It should be understood that the above description of the preferred embodiments is given for clarity and not for any purpose of limitation, and that various changes, substitutions and alterations can be made herein without departing from the spirit and scope of the invention as defined by the appended claims.
Claims (5)
1. A depression angle face recognition method based on an axial attention weight distribution network is characterized by comprising the following steps:
step 1: carrying out feature extraction on a depression angle face image to be recognized;
collectingnThe method comprises the steps that face images with different angles and different resolutions are displayed, feature extraction is carried out on face image sequences with different angles and different resolutions, a group of normalized feature vectors are generated, and the normalized feature vectors respectively represent the features of the face images with different resolutions and different angles which may appear in a section of monitoring video; wherein,nis a preset value;
the method comprises the following steps of (1) extracting features of a face image by adopting a CNN network; the CNN network adopts an ArcFace backbone network and consists of ResNet 50; after the image is subjected to feature extraction, generating a feature vector with dimensions of 1 multiplied by 512 after standardization; thus, extract outnFeatures of the input image are obtainednA feature vectorF 1 ~F n ;
And 2, step: inputting the depression angle face image after feature extraction into a cascade axial attention weight distribution module for feature aggregation;
the cascade axial attention weight distribution module uses a cascade axial attention block with nonlinear transfer inside and is responsible for obtaining a weight matrix; the cascade axial attention block comprises a transverse attention block and a longitudinal attention block, wherein the transverse attention block consists of two kernel convolutions and a nonlinear conversion function, the convolution kernels are realized through an FC layer, the nonlinear conversion is realized through a Sigmoid activation function, a self-attention weight parameter of the feature is obtained, and the weight is distributed to the feature; the vertical attention block is composed of two Conv layers and a Softmax function and generates the weight of the weight relation between vertical features; multiplying the two axial weights to obtain an attention parameter in a matrix form, and distributing attention to the multi-frame image to obtain a final polymerization vector;
and step 3: and calculating similarity by using the aggregated features through a cosine similarity function, and performing feature matching on different faces to realize face recognition.
2. The method for identifying a dip angle face based on an axial attention weight distribution network as claimed in claim 1, wherein the feature aggregation process in step 2 comprises the following sub-steps:
step 2.1: inputting the features into a cascade axial attention weight distribution module to perform weight self-adaptive distribution, wherein the weight self-adaptive distribution comprises a weight vector of each feature map and a weight vector among the features to form an axial weight matrix; wherein, the weight vector of each characteristic graph is marked as a transverse weight vector, and the weight vector between the characteristics is marked as a longitudinal weight vector;
step 2.2: and carrying out weighted aggregation by using the axial weight to obtain the final aggregation characteristic vector with stronger discriminability.
3. The dip angle face recognition method based on the axial attention weight distribution network as claimed in claim 2, wherein the step 2.1 is implemented by the following steps:
step 2.1.1: extracting a transverse attention parameter through a transverse attention weight distribution module, wherein the transverse attention weight distribution module comprises two FC layers and a Sigmoid function, and obtaining a self-attention weight A of each featureiWherein A isiSize and FiConsistently, the weight assignment results inside each feature are represented:
wherein F is extracted in the step 1nDerived from features of the input imagenA feature vector;
step 2.1.2: performing self-attention fusion on each feature by using the self-attention weight Ai to obtain a transverse weight distribution result S:
step 2.1.3: longitudinal weight distribution is carried out through a longitudinal attention weight distribution module, weight among all the characteristics is distributed in the longitudinal process, and the longitudinal attention weight distribution module comprises two Conv layers and one Softmax function to obtainnWeight W between features representing different picture featuresiW represents the vertical weight assignment:
4. the dip angle face recognition method based on the axial attention weight distribution network according to claim 3, characterized in that: in step 2.2, the features are aggregated using the obtained weight assignment results, and the aggregated features r are represented as follows:
the symbol "", indicates a dot product operation.
5. A depression angle face recognition system based on an axial attention weight distribution network is characterized by comprising the following modules:
the characteristic extraction module is used for extracting the characteristics of the facial image to be identified;
collection ofnThe method comprises the steps of opening human face images with different angles and different resolutions, carrying out feature extraction on human face image sequences with different angles and different resolutions, generating a group of normalized feature vectors, and respectively representing the features of the human face images with different resolutions and different angles which may appear in a section of monitoring video; wherein,nis a preset value;
the method comprises the following steps of (1) extracting features of a face image by adopting a CNN (convolutional neural network); the CNN network adopts an ArcFace backbone network and consists of ResNet 50; after the image passes through a feature extraction module, generating a feature vector with dimensions of 1 multiplied by 512 after standardization; thus, extract outnFeatures of the input image are obtainednA feature vectorF 1 ~F n ;
The characteristic aggregation module is used for inputting the depression angle face image subjected to characteristic extraction into the cascade axial attention weight distribution module for characteristic aggregation;
the cascade axial attention weight distribution module uses a cascade axial attention block with nonlinear transfer inside and is responsible for obtaining a weight matrix; the cascade axial attention block comprises a transverse attention block and a longitudinal attention block, wherein the transverse attention block consists of two kernel convolutions and a nonlinear conversion function, the convolution kernels are realized through an FC layer, the nonlinear conversion is realized through a Sigmoid activation function, a self-attention weight parameter of the feature is obtained, and the weight is distributed to the feature; the vertical attention block is composed of two Conv layers and a Softmax function and generates the weight of the weight relation between vertical features; multiplying the two axial weights to obtain an attention parameter in a matrix form, and distributing attention to the multi-frame image to obtain a final polymerization vector;
and the face recognition module calculates similarity through a cosine similarity function by using the aggregated features, and performs feature matching on different faces to realize face recognition.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210534034.3A CN114639156B (en) | 2022-05-17 | 2022-05-17 | Depression angle face recognition method and system based on axial attention weight distribution network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210534034.3A CN114639156B (en) | 2022-05-17 | 2022-05-17 | Depression angle face recognition method and system based on axial attention weight distribution network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114639156A CN114639156A (en) | 2022-06-17 |
CN114639156B true CN114639156B (en) | 2022-07-22 |
Family
ID=81952804
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210534034.3A Active CN114639156B (en) | 2022-05-17 | 2022-05-17 | Depression angle face recognition method and system based on axial attention weight distribution network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114639156B (en) |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111539343A (en) * | 2020-04-26 | 2020-08-14 | 安徽皖仪科技股份有限公司 | Black smoke vehicle detection method based on convolution attention network |
CN113065460A (en) * | 2021-03-31 | 2021-07-02 | 吉林农业大学 | Establishment method of pig face facial expression recognition framework based on multitask cascade |
CN113469942A (en) * | 2021-06-01 | 2021-10-01 | 天津大学 | CT image lesion detection method |
CN113706404A (en) * | 2021-08-06 | 2021-11-26 | 武汉大学 | Depression angle human face image correction method and system based on self-attention mechanism |
CN113822203A (en) * | 2021-09-26 | 2021-12-21 | 中国民用航空飞行学院 | Face recognition device and method based on reinforcement learning and deep convolutional neural network |
CN113902789A (en) * | 2021-09-24 | 2022-01-07 | Oppo广东移动通信有限公司 | Image feature processing method, depth image generating method, depth image processing apparatus, depth image generating medium, and device |
EP3979207A1 (en) * | 2020-10-02 | 2022-04-06 | Google LLC | Conditional axial transformer layers for high-fidelity image transformation |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111739075B (en) * | 2020-06-15 | 2024-02-06 | 大连理工大学 | Deep network lung texture recognition method combining multi-scale attention |
-
2022
- 2022-05-17 CN CN202210534034.3A patent/CN114639156B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111539343A (en) * | 2020-04-26 | 2020-08-14 | 安徽皖仪科技股份有限公司 | Black smoke vehicle detection method based on convolution attention network |
EP3979207A1 (en) * | 2020-10-02 | 2022-04-06 | Google LLC | Conditional axial transformer layers for high-fidelity image transformation |
CN113065460A (en) * | 2021-03-31 | 2021-07-02 | 吉林农业大学 | Establishment method of pig face facial expression recognition framework based on multitask cascade |
CN113469942A (en) * | 2021-06-01 | 2021-10-01 | 天津大学 | CT image lesion detection method |
CN113706404A (en) * | 2021-08-06 | 2021-11-26 | 武汉大学 | Depression angle human face image correction method and system based on self-attention mechanism |
CN113902789A (en) * | 2021-09-24 | 2022-01-07 | Oppo广东移动通信有限公司 | Image feature processing method, depth image generating method, depth image processing apparatus, depth image generating medium, and device |
CN113822203A (en) * | 2021-09-26 | 2021-12-21 | 中国民用航空飞行学院 | Face recognition device and method based on reinforcement learning and deep convolutional neural network |
Non-Patent Citations (4)
Title |
---|
AN IMAGE IS WORTH 16X16 WORDS:TRANSFORMERS FOR IMAGE RECOGNITION AT SCALE;Alexey Dosovitskiy等;《arXiv》;20210603;第1-22页 * |
Axial-DeepLab: Stand-Alone Axial-Attention for Panoptic Segmentation;Huiyu Wang等;《Computer Vision – ECCV 2020》;20201029;第108-126页 * |
Realistic frontal face reconstruction using coupled complementarity of far-near-sighted face images;Kangli Zeng等;《Pattern Recognition》;20220427;第1-11页 * |
深度多尺度融合注意力残差人脸表情识别网络;高涛 等;《智能***学报》;20220331;第17卷(第2期);第393-401页 * |
Also Published As
Publication number | Publication date |
---|---|
CN114639156A (en) | 2022-06-17 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110706157B (en) | Face super-resolution reconstruction method for generating confrontation network based on identity prior | |
KR102147052B1 (en) | Emotional recognition system and method based on face images | |
CN103731583B (en) | Intelligent synthetic, print processing method is used for taking pictures | |
Liu et al. | Pose-guided R-CNN for jersey number recognition in sports | |
CN112766160A (en) | Face replacement method based on multi-stage attribute encoder and attention mechanism | |
CN111639564B (en) | Video pedestrian re-identification method based on multi-attention heterogeneous network | |
CN111241975B (en) | Face recognition detection method and system based on mobile terminal edge calculation | |
CN111639580B (en) | Gait recognition method combining feature separation model and visual angle conversion model | |
CN106529467A (en) | Group behavior identification method based on multi-feature fusion | |
CN107330397A (en) | A kind of pedestrian's recognition methods again based on large-spacing relative distance metric learning | |
CN102332095A (en) | Face motion tracking method, face motion tracking system and method for enhancing reality | |
JP5574033B2 (en) | Image recognition system, recognition method thereof, and program | |
CN111539255A (en) | Cross-modal pedestrian re-identification method based on multi-modal image style conversion | |
WO2021139171A1 (en) | Facial enhancement based recognition method, apparatus and device, and storage medium | |
CN110765841A (en) | Group pedestrian re-identification system and terminal based on mixed attention mechanism | |
CN108537196B (en) | Human behavior identification method of space-time distribution diagram generated based on motion history point cloud | |
CN113963032A (en) | Twin network structure target tracking method fusing target re-identification | |
CN114078275A (en) | Expression recognition method and system and computer equipment | |
Zhu et al. | A two-stage detector for hand detection in ego-centric videos | |
Ahmad et al. | Event-driven re-id: A new benchmark and method towards privacy-preserving person re-identification | |
CN115439884A (en) | Pedestrian attribute identification method based on double-branch self-attention network | |
CN111079585B (en) | Pedestrian re-identification method combining image enhancement with pseudo-twin convolutional neural network | |
CN114639156B (en) | Depression angle face recognition method and system based on axial attention weight distribution network | |
Xu et al. | 2d license plate recognition based on automatic perspective rectification | |
Özyurt et al. | A new method for classification of images using convolutional neural network based on Dwt-Svd perceptual hash function |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |