CN110738071A - face algorithm model training method based on deep learning and transfer learning - Google Patents
face algorithm model training method based on deep learning and transfer learning Download PDFInfo
- Publication number
- CN110738071A CN110738071A CN201810787805.3A CN201810787805A CN110738071A CN 110738071 A CN110738071 A CN 110738071A CN 201810787805 A CN201810787805 A CN 201810787805A CN 110738071 A CN110738071 A CN 110738071A
- Authority
- CN
- China
- Prior art keywords
- training
- layer
- model
- learning
- neural network
- 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.)
- Pending
Links
- 238000012549 training Methods 0.000 title claims abstract description 51
- 238000013526 transfer learning Methods 0.000 title claims abstract description 21
- 238000000034 method Methods 0.000 title claims abstract description 16
- 238000013135 deep learning Methods 0.000 title claims abstract description 14
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 24
- 230000004913 activation Effects 0.000 claims description 6
- 238000011156 evaluation Methods 0.000 claims description 6
- 238000011176 pooling Methods 0.000 claims description 6
- 238000013528 artificial neural network Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 2
- 230000001815 facial effect Effects 0.000 claims 1
- 238000012360 testing method Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/161—Detection; Localisation; Normalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/29—Graphical models, e.g. Bayesian networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/172—Classification, e.g. identification
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Human Computer Interaction (AREA)
- Oral & Maxillofacial Surgery (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention provides face algorithm model training methods based on deep learning and transfer learning, which are characterized in that a convolutional neural network is trained on a pure portrait dataset of a large-capacity sample to obtain a pre-training model, then, on the basis, the final model is obtained by retraining a portrait dataset (a data set of a portrait and an identity card photo) through transfer learning, and the problem of how to use fewer portrait training samples to train a high-accuracy portrait recognition model is effectively solved.
Description
[ technical field ] A method for producing a semiconductor device
The invention relates to the field of face image recognition, in particular to testimony recognition training methods based on deep learning and transfer learning.
[ background of the invention ]
The human face recognition is branches of image recognition, two human face photos can be fully automatically verified by using a human face verification algorithm, and whether the two human face photos are the same people or not can be judged.
[ summary of the invention ]
In order to solve the problems, the invention provides face algorithm model training methods based on deep learning and transfer learning, a convolutional neural network is trained on a pure portrait dataset of a large-capacity sample to obtain a pre-training model, then on the basis, the transfer learning is carried out, and retraining is carried out on a testimony dataset (a portrait and an identification card photo dataset) to obtain a final model, so that the problem of how to use fewer testimony training samples to train a testimony recognition model with high accuracy is effectively solved.
The invention discloses an face algorithm model training method based on deep learning and transfer learning, which comprises the following steps:
step 1: training a pure portrait dataset of a collected large-capacity sample by adopting a convolutional neural network to obtain a pre-training model, wherein the convolutional neural network model sequentially comprises a training data layer, a convolutional layer, an activation layer, a pooling layer, a full-link layer and a loss layer;
step 2: and (3) taking the pre-training model in the step (1) as a training starting point, copying network parameters before the full-connection layer in the step (1) by adopting transfer learning, and training the full-connection layer and the loss layer of the testimonial convolutional neural network on the testimonial data set in an important way to obtain a final model.
Because the sample of the human image data set is less, when the convolutional neural network training is directly carried out on the human image data set, a network model with good effect is difficult to train, the sample of the human image data set is more, the human image data set and the human image data set are all picture sets containing human faces, and the learning task is face recognition, so that the human image data set can be used for transfer learning. Therefore, in order to improve the accuracy of the human face identification of the testimony, a convolutional neural network is trained on a human image data set, parameters of a data layer, a convolutional layer, an activation layer and a pooling layer of a pre-training model are copied into the testimony convolutional neural network, and a full connection layer and a loss layer of the testimony convolutional neural network are trained in an important mode.
As technical solutions, the training of the data layer includes the following steps:
s1, detecting the human face, namely, inputting pictures containing the human face, and detecting the position of the human face by adopting a cascade structure and a multilayer neural network;
s2: key point positioning: according to the detected face, 5 key points are extracted by adopting a coarse-fine self-encoder network: left eye center, right eye center, nose tip, left mouth corner and right mouth corner;
s3: face preprocessing: the face images detected in step S1 are aligned using similarity transformation of 5 key points detected in step S2 with 5 given standard key points, and the aligned images have the same size.
As technical solutions, the loss function of the convolutional neural network training adopts softmax loss and center loss, and the calculation formula is as follows:
wherein,
is the total loss;
the difference between classes can be increased if the model is softmax;the method is characterized in that the method comprises the steps of learning classes of centers and reducing the intra-class distance, wherein lambda is the weight occupied by the center loss.
As technical solutions, in step 2, the learning rate of each layer before the full connection layer is set to 0 or reduced, because the parameters of each layer before the full connection layer are generalization-capable, with little or no need of retraining, the full connection layer and the loss layer need to be retrained to extract the specific features of the testimonial data set.
As technical solutions, the pre-training model obtained according to step 1 is compared with an evaluation set of face recognition international authority, the evaluation set includes LFW and megaface, and the accuracy of the pre-training model can be further guaranteed by comparing the pre-training model with the evaluation set of face recognition international authority.
In conclusion, the invention has the advantages that based on deep learning and transfer learning, convolutional neural network training is carried out on the pure portrait dataset of the large-capacity sample, and on the basis of the obtained training result, the testimony recognition model with higher accuracy is obtained by carrying out fine adjustment on the testimony dataset, thereby solving the problems that the testimony dataset has fewer samples and can not be directly applied to deep learning.
[ description of the drawings ]
FIG. 1 is a general flow chart of example 1 of the present invention
FIG. 2 is a flowchart of convolutional neural network training of human image data set according to embodiment 1 of the present invention
FIG. 3 is a flowchart of transfer learning according to embodiment 1 of the present invention
[ detailed description ] embodiments
Example 1
As shown in fig. 1, the face algorithm model training methods based on deep learning and transfer learning provided by the present invention firstly train a convolutional neural network on a pure human image data set of a large-capacity sample, as shown in fig. 2, including the following steps:
1. face detection, pictures containing faces are input, and the positions of the faces are detected by adopting a cascade structure and a multilayer neural network;
2. key point positioning: according to the detected face, 5 key points are extracted by adopting a coarse-fine self-encoder network: left eye center, right eye center, nose tip, left mouth corner and right mouth corner;
3. face preprocessing: aligning the face image detected in the step 1 by using the similarity transformation of the 5 key points detected in the step 2 and 5 given standard key points, wherein the aligned images have the same size;
4. network training: training is carried out by adopting a convolutional neural network, and softmax loss and center loss are adopted as loss functions, and the formula is as follows:
wherein,
is the total loss that is to be expected,
is softmax, the inter-class gap can be increased,
is centerloss, it is possible to learn the centers of classes and reduce the intra-class distance, and λ is the weight taken by the center loss.
The convolutional neural network simultaneously comprises a data layer, a convolutional layer, an activation layer, a pooling layer, a full-link layer and a loss layer, wherein the data layer can be obtained through the steps 1-3, the convolutional layer is used for convolving images so as to implicitly extract features from training data, the activation layer is added with nonlinear factors so as to improve the expression capacity of the convolutional neural network, the pooling layer is used for downsampling a feature mapping surface and mainly reduces dimensions, the full-link layer plays a role of a classifier in the whole convolutional neural network, the features are obtained in the full-link layer, and the features are groups of fixed-length numbers.
After the convolutional neural network of the portrait data set is trained, an evaluation set LFW for identifying international authority by a human face is used for comparison and test, and a pre-training model after accuracy test is used for training the testimony data set.
Because the portrait data set and the testimony data set are all picture sets containing human faces, and the learning task is human face recognition, the transfer learning can be used, and the transfer learning process is shown in fig. 3, and the specific method is as follows:
A. inputting a testimony data set, and taking a pre-training model of the portrait data set as a training starting point;
B. copying parameters of a data layer, a convolutional layer, an activation layer and a pooling layer of a pre-training model into a testimonial convolutional neural network, and setting the learning rate of the convolutional layer to be 0;
C. and training a full connection layer and a loss layer of the testimonial convolutional neural network to obtain a final model.
Claims (5)
1, a face algorithm model training method based on deep learning and transfer learning, which is characterized by comprising the following steps:
step 1: training a pure portrait dataset of a collected large-capacity sample by adopting a convolutional neural network to obtain a pre-training model, wherein the convolutional neural network model sequentially comprises a training data layer, a convolutional layer, an activation layer, a pooling layer, a full-link layer and a loss layer;
step 2: and (3) taking the pre-training model in the step (1) as a training starting point, copying network parameters before the full-connection layer in the step (1) by adopting transfer learning, and training the full-connection layer and the loss layer of the testimonial convolutional neural network on the testimonial data set in an important way to obtain a final model.
2. The method for training facial algorithm models based on deep learning and transfer learning of claim 1, wherein the training of the data layer comprises the following steps:
s1, detecting the human face, namely, inputting pictures containing the human face, and detecting the position of the human face by adopting a cascade structure and a multilayer neural network;
s2: key point positioning: according to the detected face, 5 key points are extracted by adopting a coarse-fine self-encoder network: left eye center, right eye center, nose tip, left mouth corner and right mouth corner;
s3: face preprocessing: the face images detected in step S1 are aligned using similarity transformation of 5 key points detected in step S2 with 5 given standard key points, and the aligned images have the same size.
3. The deep learning and transfer learning-based face algorithm model training method of claim 1, wherein the loss function of convolutional neural network training is softmax loss and center loss, and the calculation formula is as follows:
wherein,is the total loss;the difference between classes can be increased if the model is softmax;
is centerloss, it is possible to learn the centers of classes and reduce the intra-class distance, and λ is the weight taken by the center loss.
4. The human face algorithm model training methods based on deep learning and transfer learning of claim 1, wherein in the step 2, the learning rate of each layer before the fully connected layer is set to 0 or reduced.
5. The training method of human face algorithm models based on deep learning and transfer learning according to claim 1, wherein the pre-training model obtained according to step 1 is compared with an evaluation set of human face recognition international authority, and the accuracy of the pre-training model is tested, wherein the evaluation set comprises LFW and MegaFace.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810787805.3A CN110738071A (en) | 2018-07-18 | 2018-07-18 | face algorithm model training method based on deep learning and transfer learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810787805.3A CN110738071A (en) | 2018-07-18 | 2018-07-18 | face algorithm model training method based on deep learning and transfer learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110738071A true CN110738071A (en) | 2020-01-31 |
Family
ID=69233561
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810787805.3A Pending CN110738071A (en) | 2018-07-18 | 2018-07-18 | face algorithm model training method based on deep learning and transfer learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110738071A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109376692A (en) * | 2018-11-22 | 2019-02-22 | 河海大学常州校区 | Migration convolution neural network method towards facial expression recognition |
| CN111898454A (en) * | 2020-07-02 | 2020-11-06 | 中国地质大学（武汉） | Weight binarization neural network and transfer learning human eye state detection method and device |
| CN112395986A (en) * | 2020-11-17 | 2021-02-23 | 广州像素数据技术股份有限公司 | Face recognition method for quickly migrating new scene and preventing forgetting |
| CN113436064A (en) * | 2021-08-26 | 2021-09-24 | 北京世纪好未来教育科技有限公司 | Method and equipment for training detection model of key points of target object and detection method and equipment |
| WO2021203718A1 (en) * | 2020-04-10 | 2021-10-14 | 嘉楠明芯(北京)科技有限公司 | Method and system for facial recognition |
| CN114187173A (en) * | 2021-12-15 | 2022-03-15 | 北京欧珀通信有限公司 | Model training method, image processing method and device, electronic device and medium |
| WO2023061169A1 (en) * | 2021-10-11 | 2023-04-20 | 北京字节跳动网络技术有限公司 | Image style migration method and apparatus, image style migration model training method and apparatus, and device and medium |
| CN113505740B (en) * | 2021-07-27 | 2023-10-10 | 北京工商大学 | Face recognition method based on transfer learning and convolutional neural network |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106295584A (en) * | 2016-08-16 | 2017-01-04 | 深圳云天励飞技术有限公司 | Depth migration study is in the recognition methods of crowd's attribute |
| CN106650660A (en) * | 2016-12-19 | 2017-05-10 | 深圳市华尊科技股份有限公司 | Vehicle type recognition method and terminal |
| CN106951867A (en) * | 2017-03-22 | 2017-07-14 | 成都擎天树科技有限公司 | Face identification method, device, system and equipment based on convolutional neural networks |
| CN107103281A (en) * | 2017-03-10 | 2017-08-29 | 中山大学 | Face identification method based on aggregation Damage degree metric learning |
| CN107832700A (en) * | 2017-11-03 | 2018-03-23 | 全悉科技（北京）有限公司 | A kind of face identification method and system |
| CN107886064A (en) * | 2017-11-06 | 2018-04-06 | 安徽大学 | A kind of method that recognition of face scene based on convolutional neural networks adapts to |
| CN108009528A (en) * | 2017-12-26 | 2018-05-08 | 广州广电运通金融电子股份有限公司 | Face authentication method, device, computer equipment and storage medium based on Triplet Loss |
| CN108133238A (en) * | 2017-12-29 | 2018-06-08 | 国信优易数据有限公司 | A kind of human face recognition model training method and device and face identification method and device |
| CN108182427A (en) * | 2018-01-30 | 2018-06-19 | 电子科技大学 | A kind of face identification method based on deep learning model and transfer learning |
| CN108280426A (en) * | 2018-01-23 | 2018-07-13 | 深圳极视角科技有限公司 | Half-light source expression recognition method based on transfer learning and device |
-
2018
- 2018-07-18 CN CN201810787805.3A patent/CN110738071A/en active Pending
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106295584A (en) * | 2016-08-16 | 2017-01-04 | 深圳云天励飞技术有限公司 | Depth migration study is in the recognition methods of crowd's attribute |
| CN106650660A (en) * | 2016-12-19 | 2017-05-10 | 深圳市华尊科技股份有限公司 | Vehicle type recognition method and terminal |
| CN107103281A (en) * | 2017-03-10 | 2017-08-29 | 中山大学 | Face identification method based on aggregation Damage degree metric learning |
| CN106951867A (en) * | 2017-03-22 | 2017-07-14 | 成都擎天树科技有限公司 | Face identification method, device, system and equipment based on convolutional neural networks |
| CN107832700A (en) * | 2017-11-03 | 2018-03-23 | 全悉科技（北京）有限公司 | A kind of face identification method and system |
| CN107886064A (en) * | 2017-11-06 | 2018-04-06 | 安徽大学 | A kind of method that recognition of face scene based on convolutional neural networks adapts to |
| CN108009528A (en) * | 2017-12-26 | 2018-05-08 | 广州广电运通金融电子股份有限公司 | Face authentication method, device, computer equipment and storage medium based on Triplet Loss |
| CN108133238A (en) * | 2017-12-29 | 2018-06-08 | 国信优易数据有限公司 | A kind of human face recognition model training method and device and face identification method and device |
| CN108280426A (en) * | 2018-01-23 | 2018-07-13 | 深圳极视角科技有限公司 | Half-light source expression recognition method based on transfer learning and device |
| CN108182427A (en) * | 2018-01-30 | 2018-06-19 | 电子科技大学 | A kind of face identification method based on deep learning model and transfer learning |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109376692A (en) * | 2018-11-22 | 2019-02-22 | 河海大学常州校区 | Migration convolution neural network method towards facial expression recognition |
| WO2021203718A1 (en) * | 2020-04-10 | 2021-10-14 | 嘉楠明芯(北京)科技有限公司 | Method and system for facial recognition |
| CN111898454A (en) * | 2020-07-02 | 2020-11-06 | 中国地质大学（武汉） | Weight binarization neural network and transfer learning human eye state detection method and device |
| CN112395986A (en) * | 2020-11-17 | 2021-02-23 | 广州像素数据技术股份有限公司 | Face recognition method for quickly migrating new scene and preventing forgetting |
| CN112395986B (en) * | 2020-11-17 | 2024-04-26 | 广州像素数据技术股份有限公司 | Face recognition method capable of quickly migrating new scene and preventing forgetting |
| CN113505740B (en) * | 2021-07-27 | 2023-10-10 | 北京工商大学 | Face recognition method based on transfer learning and convolutional neural network |
| CN113436064A (en) * | 2021-08-26 | 2021-09-24 | 北京世纪好未来教育科技有限公司 | Method and equipment for training detection model of key points of target object and detection method and equipment |
| WO2023061169A1 (en) * | 2021-10-11 | 2023-04-20 | 北京字节跳动网络技术有限公司 | Image style migration method and apparatus, image style migration model training method and apparatus, and device and medium |
| CN114187173A (en) * | 2021-12-15 | 2022-03-15 | 北京欧珀通信有限公司 | Model training method, image processing method and device, electronic device and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110738071A (en) | face algorithm model training method based on deep learning and transfer learning | |
| William et al. | Face recognition using facenet (survey, performance test, and comparison) | |
| CN107145842B (en) | Face recognition method combining LBP characteristic graph and convolutional neural network | |
| US9514356B2 (en) | Method and apparatus for generating facial feature verification model | |
| CN112232117A (en) | Face recognition method, face recognition device and storage medium | |
| CN109145717B (en) | Face recognition method for online learning | |
| CN102938065B (en) | Face feature extraction method and face identification method based on large-scale image data | |
| CN108427921A (en) | A kind of face identification method based on convolutional neural networks | |
| CN103218609B (en) | A kind of Pose-varied face recognition method based on hidden least square regression and device thereof | |
| CN108182397B (en) | Multi-pose multi-scale human face verification method | |
| CN109800643A (en) | A kind of personal identification method of living body faces multi-angle | |
| CN108563999A (en) | A kind of piece identity's recognition methods and device towards low quality video image | |
| CN108564040B (en) | Fingerprint activity detection method based on deep convolution characteristics | |
| CN104700089A (en) | Face identification method based on Gabor wavelet and SB2DLPP | |
| CN110796101A (en) | Face recognition method and system of embedded platform | |
| CN106874877A (en) | A kind of combination is local and global characteristics without constraint face verification method | |
| CN108549883A (en) | A kind of face recognition methods again | |
| CN107392191B (en) | Method for judging identity, device and electronic equipment | |
| Prabhavathi et al. | A smart technique for attendance system to recognize faces through parallelism | |
| CN103745242A (en) | Cross-equipment biometric feature recognition method | |
| WO2015131710A1 (en) | Method and device for positioning human eyes | |
| CN111950452A (en) | Face recognition method | |
| Salah et al. | Recognize Facial Emotion Using Landmark Technique in Deep Learning | |
| Tong et al. | Research on face recognition method based on deep neural network | |
| CN110287973B (en) | Image feature extraction method based on low-rank robust linear discriminant analysis |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20200131 |