CN106778589A - A kind of masked method for detecting human face of robust based on modified LeNet - Google Patents
A kind of masked method for detecting human face of robust based on modified LeNet Download PDFInfo
- Publication number
- CN106778589A CN106778589A CN201611127956.3A CN201611127956A CN106778589A CN 106778589 A CN106778589 A CN 106778589A CN 201611127956 A CN201611127956 A CN 201611127956A CN 106778589 A CN106778589 A CN 106778589A
- Authority
- CN
- China
- Prior art keywords
- mlenet
- masked
- face
- model
- lenet
- 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
- 241000282414 Homo sapiens Species 0.000 title claims abstract description 26
- 238000000034 method Methods 0.000 title claims abstract description 19
- 238000012549 training Methods 0.000 claims abstract description 53
- 238000001514 detection method Methods 0.000 claims abstract description 38
- 238000005516 engineering process Methods 0.000 claims abstract description 14
- 208000032443 Masked facies Diseases 0.000 claims abstract 3
- 230000005764 inhibitory process Effects 0.000 claims description 3
- 230000001629 suppression Effects 0.000 abstract description 5
- 238000012360 testing method Methods 0.000 abstract description 4
- 230000008859 change Effects 0.000 abstract description 3
- 230000007613 environmental effect Effects 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 15
- 238000013527 convolutional neural network Methods 0.000 description 5
- 238000011161 development Methods 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000011478 gradient descent method Methods 0.000 description 4
- 238000010200 validation analysis Methods 0.000 description 4
- 230000006872 improvement Effects 0.000 description 3
- 230000000007 visual effect Effects 0.000 description 3
- 230000006399 behavior Effects 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 230000000873 masking effect Effects 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 238000012805 post-processing Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 230000003190 augmentative effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000014509 gene expression Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
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
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24137—Distances to cluster centroïds
- G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
A kind of masked method for detecting human face of robust based on modified LeNet, is related to masked Face datection.Comprise the following steps：1) by the original training picture of flip horizontal, training data is expanded；2) by changing the structure of traditional LeNet models, new MLeNet models are proposed, is allowed to be adapted to the test problems of the masked mankind, specific method can be：Adjustment convolution kernel size and characteristic pattern number, in addition, the nodes 10 for changing original output layer are 2, make it suitable for 2 classification problems of mankind's detection；3) the parameter pre-training MLeNet structures in original LeNet models are borrowed, and finely tunes MLeNet models, obtain being suitable for the detector of masked face；4) position of masked man's face is accurately positioned out with reference to sliding window and non-maximization suppression technology.Masked man's face can be accurately detected, and it is at random in background, under the disturbed condition such as environmental change, the model still has stronger robustness.
Description
Technical Field
The invention relates to mask face detection, in particular to a robust mask face detection method based on an improved LeNet.
Background
With the development of society, the improvement of science and technology and the popularization of multimedia technology, more and more people upload various network videos on the network, wherein the activities include that many criminals attempt to use multimedia channels to start to spread violent terrorism videos, and the behaviors influence the stable development of society to a certain extent. If terrorists can be quickly and accurately located in massive video frames, human resources are greatly reduced and social stability is maintained.
As a basic requirement for the management of a large-scale video library, accurately retrieving an violently terrorist video frame in possession of a terrorist plays a significant role in the overall social stability. How to accurately define in a given video frame there is a terrorist problem, which is a difficult problem because terrorists behave in a wide variety of forms. Since terrorists are generally masked, in the present invention, terrorists are considered to be persons with masking characteristics. Mask human face detection is a special task of face detection, and is different from the traditional face detection technology in that more challenges are faced. On one hand, the masked human face detection includes the influence conditions of posture change, illumination and the like which cannot be processed by the traditional face detection technology. On the other hand, the face of the masked person is severely shielded, so that the normal structure of the original face is greatly lost, and the traditional algorithm fails to detect the face of the masked person.
Currently, a number of face detection techniques rely on manually set features, such as: widely used Fisherface [1], a cascade classifier based on Haar-like features [2], and an AdaBoost detector based on Gabor-like high-dimensional features [3 ]. Because the manually set features require a large number of training samples and the mask person loses the complete face structure, the manually designed features cannot accurately represent the face structure of the mask person, and finally the methods cannot accurately detect the face of the mask person. Recently, template-based (exemplar-based) face detection methods [4] have shown good results, mainly because the huge template database covers all possible face visual changes (visual variations), including changes in occlusion, illumination, face pose, etc., but this method requires a large amount of template data sets and is prone to false alarm (false alarm) results in highly scattered background situations. In order to reduce the number of required templates, document [5] proposes an effective template face detection method based on lifting. The method can further improve the human face detection rate, accelerate the detection process, and greatly save the memory overhead by using a mode of combining discriminant training and effectiveness as a weak classifier.
In recent years, due to the rise of deep learning, a Convolutional Neural Network (CNN) with strong GPU computing power has made a great breakthrough in the field of human faces, such as LFW [6] [7] [8 ]. In particular, convolutional networks are able to automatically learn valid feature representations through training samples. In the 2012 Large Scale Recognition competition (Large Scale Visual Recognition Challenge), document [9] made a breakthrough progress with deep convolutional neural networks. In addition, in order to further process the situation with only a small number of training samples, document [10] introduces weights for pre-training the initialized deep network, so as to speed up the convergence of the network and obtain a better local solution. Document [11] proposes a LeNet model, which shows good performance in handwritten character recognition. With the development of these deep learning techniques, face detection methods based on deep learning become possible.
Reference documents:
[1]H.J.P.Belhumeur P N,K.D.J.Eigenfaces vs.fisherfaces:Recognitionusing class specific linear projection.IEEE Transactions on Pattern Analysisand Machine Intelligence,1997,19(7):711-720.
[2]P.Viola,M.Jones,Rapid object detection using a boosted cascade ofsimple features.in Proceedings of CVPR,2001.
[3]C.Liu,H.Wechsler,Gabor feature based classification using theenhanced fisher linear discriminant model for face recognition.IEEETransactions on Image Processing,2002,11(4):467-476.
[4]X.Shen,Z.Lin,J.Brandt,et al.Detecting and aligning faces by imageretrieval.in Proceedings of CVPR,2013:3460-3467.
[5]H.Li,Z.Lin,J.Brandt,et al.Efficient boosted exemplar-based facedetection.In Proceedings of CVPR,2014:1843-1850.
[6]X.W.Yi Sun,X.Tang.Deep learning face representation frompredicting10,000classes.in Proceedings of CVPR,2014:1891-1898.
[7]Y.Sun,X.Wang,X.Tang.Deeply learned face representations aresparse,selective,and robust.arXiv preprint arXiv:1412.1265.
[8]Y.Sun,X.Wang,X.Tang.Hybrid deep learning for face verification.inProceedings of ICCV,2013:1489-1496.
[9]A.Krizhevsky,I.Sutskever,G.E.Hinton.Imagenet classification withdeep convolutional neural networks.in Proceedings of NIPS,2012:1097-1105.
[10]G.E.Hinton,R.R.Salakhutdinov.Reducing the dimensionality of datawith neural networks.Science,2006,313:504-507.
[11]Y.LeCun,L.Bottou,Y.Bengio,et al.Gradient-based learning appliedto document recognition.Proceedings of the IEEE,1998,86(11):2278-2324.
disclosure of Invention
Aiming at the characteristics that training samples are few and the complete structural features of a mask cannot be obtained, the invention provides a robust mask face detection method based on an improved LeNet, which can quickly and accurately position the face position of the mask by introducing pre-training and fine-tuning (pre-training and fine-tuning) and other means and combining a sliding window method for MLeNet.
The invention comprises the following steps:
1) expanding training data by horizontally turning over the original training picture;
2) by modifying the structure of the traditional LeNet model, a new MLeNet model is provided to adapt to the detection problem of masked human beings, and the specific method can be as follows: adjusting the size of a convolution kernel and the number of characteristic graphs, and changing the number 10 of nodes of an original output layer to be 2 so as to be suitable for the 2-classification problem of human detection;
3) pre-training an MLeNet structure by using parameters in an original LeNet model, and finely adjusting the MLeNet model to obtain a detector suitable for a mask face;
4) and the position of the face of the masked person is accurately positioned by combining a sliding window and a non-maximization inhibition technology.
The invention has the following outstanding advantages:
the invention provides a new MLeNet model by modifying the size of convolution kernel (convolution filter), the number of characteristic maps (feature maps) and the number of nodes of a full connection layer on the basis of an original LeNet model. Meanwhile, the performance of the MLeNet is further improved by means of expanding training samples, combining pre-training, fine adjustment and the like. And finally, accurately positioning the face position of the masked person by combining a sliding window and non-maximum suppression (non-maximum suppression). In the present invention, the requirement for the device is low, only one 8G U disk is needed for storing the data set for training the MLeNet model, and in addition, a high performance CPU is needed for calculating various convolution calculations in the MLeNet model.
The invention has the following technical effects:
the model provided by the invention can accurately detect the face of a masked person by using the modified LeNet model, providing a new MLeNet model, utilizing technologies such as pre-training, fine adjustment and data expansion and introducing some post-processing technologies, and still has stronger robustness under the interference conditions of scattered background, environmental change and the like.
The MLeNet model can effectively solve the problem of model overfitting caused by the problem of small samples, can accurately position the face position of a mask person in a natural environment, and has a great amount of application prospects in the fields of video monitoring, public safety and the like. The invention establishes an MLeNet model, which modifies the original LeNet model, so that the model is more suitable for the detection of the face of a masked person. Under the condition of less training samples, the model is trained to easily cause the over-fitting phenomenon, so the over-fitting problem is solved and the classification accuracy of the MLeNet model is improved by expanding a training data set and combining the technologies of pre-training, fine adjustment and the like. The use of post-processing methods, such as non-maximum suppression, makes the detection of the face of the masked person more accurate.
Drawings
Fig. 1 is a general flowchart of specific mask face detection.
Fig. 2 is a modified convolutional neural network MLeNet model: the MLeNet output layer has only two nodes, and all convolution layers have smaller convolution kernel size and each layer has larger feature map number.
Fig. 3 is a LeNet loss function value (function loss value including training and validation phases).
Fig. 4 is a graph of the LeNet classification error rate (classification error rate including positive and negative samples).
Fig. 5 shows the MLeNet loss function values without pre-training and fine-tuning (including the function loss values during the training and validation phases).
FIG. 6 shows the classification error rate of MLeNet (including the classification error rate of positive and negative samples) without pre-training and fine-tuning.
Fig. 7 shows the MLeNet loss function values (including the function loss values during the training and validation phases) with pre-training and fine-tuning.
FIG. 8 is a graph of the MLeNet classification error rate (including the classification error rate of positive and negative samples) with pre-training and fine tuning.
Fig. 9 shows partial results of masked terrorist face detection (the face region of the masked person is processed by a mosaic to protect privacy).
Detailed Description
The invention aims to provide an MLeNet model aiming at the problems of few training samples and improvement of the traditional manual adjustment of human face characteristics, obtains an accurate robust human face model through simple means of sample extension, pre-training, fine adjustment and the like, and obtains a fast, robust and accurate human face detector by combining a sliding window and a non-maximization inhibition method. The specific algorithm flow is shown in fig. 1. Each module is specifically as follows:
1. augmenting data sets
The training and testing data set used by the invention is formed by combining a plurality of key frames in the department violence and terrorism video provided by the ministry of public security. A total of 1140 pictures were included, of which 240 positive samples (i.e. containing a mask face), 900 negative samples (i.e. containing no mask face), and the experiment was conducted by randomly selecting 150 positive samples and 750 negative samples as a training set (training set), 50 positive samples and 50 negative samples as a verification set (validation set), and leaving 140 pictures as a test set (test set). In consideration of the special symmetric information of the human face, the invention utilizes a horizontal inversion (horizontal inversion) technology to expand the original data set by two times.
2. MLeNet model
The MLeNet model is an improved LeNet model. The LeNet model has 5 layers in total, 3 convolutional layers (convolutional layers) and 2 fully connected layers (fully connected layers), and the convolutional layers contain convolution and downsampling operations. Firstly, considering whether a problem of a masked human face exists, which is a two-classification problem, the number of nodes of the last fully-connected layer is modified to be 2 from the original 10, and the size of a convolution kernel in the original LeNet is reduced to 3 multiplied by 3, but the number of feature maps of each layer is increased. Specifically, the number of nodes changing the first full link layer (FC4) is increased from 84 to 500. Each layer of information for the MLeNet and LeNet models is detailed in table 1, and the final MLeNet model is shown in fig. 2.
MLeNet and LeNet models see table 1: each model contains 3 convolutional layers and 2 fully-connected layers, and the detailed layer parameters of each model are listed in the last two rows, where the convolutional kernel size "num × size × size", the convolutional kernel shift interval "st.", the space-filling "pad", and the maximum pool factor.
TABLE 1
Let N training samples beWherein the label yiIs a tag variable (in the present invention the value is 0 or 1). The final loss function is the Softmax loss function (i.e., the error of the predictor from the tag), defined as:
wherein,probability value for model output, l { yiJ is an exemplary function, which can be defined as
If the model output valueThe closer to the true tag value, the smaller the error output. w and b are respectively the weight and the deviation of each layer. Predictive tagCan be obtained by a series of w, b forward propagation. In addition, each parameter of the network can combine back-propagation (back-propagation) layer error and random gradient descent (stochastic gradient device) to update all the parameters.
Specifically, the present invention trains the MLeNet model (i.e., updates the variable w, b for each layer) using the gradient descent method, setting the batch (batch) size to 20, the momentum (momentum) to 0.9, the weight decay (weight decay) to 0.0005, the learning rate (learning rate) to 0.001, and the number of training rounds (epoch) to 100. The weight w and bias b update rules are as follows:
wherein i is an iteration index value, u and v are momentum variables,denoted as ith batch image DiThe partial derivative of the corresponding objective function to the weight w,denoted as ith batch image DiThe partial derivative of the corresponding objective function to the weight b. The updated rule shows that the variable (weight w and deviation b) of each layer is updated in a way that the target loss function moves along the direction of the local minimum value, and finally the local optimal solution is obtained. The initialized weight and offset value of the invention are directly from the trained LeNet model parameters, and MLeNet is finely adjusted by using a random gradient descent method. On a common PC (personal computer) with a 6GB memory and 1.90GHz AMD A8-4500MAPU (graphics processing Unit), the MLeNet model can be trained for 100 rounds without adopting a GPU (graphics processing Unit), and the training time only needs to take 10 min.
3. Improving the detection accuracy: pre-training, fine-tuning
The invention learns the MLeNet model through pre-training and fine-tuning means. First, the LeNet model is pre-trained with the MNIST data set, and then the MLeNet parameters are initialized through the learned LeNet parameters. Finally, the parameters of MLeNet were fine-tuned using a random gradient descent method.
4. Detecting a masked human face
By using the training MLeNet method, a masked human face detector with higher accuracy can be obtained, and whether a masked human face exists in a given window can be judged. However, the multi-scale and detection window overlap issues are not taken into account, so the present invention utilizes an image pyramid matching scheme in conjunction with non-maxima suppression to post-process such issues.
In short, in order to perform pyramid matching, target images are acquired at different positions of a multi-scale image, each sampled image is put into a trained MLeNet mask human face detector, and the MLeNet detector can generate a score value for whether a human face exists or not for each window. Then, some high-scoring sub-windows are suppressed and fused by using non-maximum values, and finally, detection is finished.
A new MLeNet model based face detection technique for masked people. MLeNet can quickly and accurately locate the position of the face of a masked person by introducing pre-training and fine-tuning (pre-training and fine-tuning) and other means and combining a sliding window method.
The specific experimental results are as follows:
with the development of society, the improvement of science and technology and the popularization of multimedia technology, more and more people upload various network videos on the network, wherein the activities include that many criminals attempt to use multimedia channels to start to spread violent terrorism videos, and the behaviors influence the stable development of society to a certain extent. If terrorists can be quickly and accurately located in massive video frames, human resources are greatly reduced and social stability is maintained. How to accurately define in a given video frame there is a terrorist problem, which is a difficult problem because terrorists behave in a wide variety of forms. Since terrorists are generally masked, in the present invention, terrorists are considered to be persons with masking characteristics. Therefore, whether the face position of the masked person can be accurately positioned is a key for judging whether terrorists exist in the video frame. Under the condition that a small number of training samples are given and a mask person cannot obtain a complete face structure, the traditional face detection technology cannot accurately position the face position of the mask person.
The face detection is an important application of the computer vision direction, the traditional face detection algorithm can accurately detect the face with the front face and no shielding face, but the good detection effect can not be obtained for the shielding condition, particularly the low resolution and the mask condition. The invention provides a new model for the detection of the face of a masked person, can obtain good performance, and can be used in the fields of video monitoring, man-machine interaction, riot and terrorist video retrieval, public safety and the like.
Fig. 3 and 4 show the performance of the LeNet model on a given set of mask face data. Fig. 5 and 6 show the performance of MLeNet without pre-training and fine-tuning, and fig. 7 and 8 show the training results of MLeNet with pre-training and fine-tuning. As can be seen from the experimental graph, the MLeNet model trained by means of pre-training, fine-tuning and the like is added, so that the mask face classification result is greatly improved.
The experimental results for the detection of the human face of the masked person in the self-created masked person dataset are shown in table 2. As can be seen from table 2, compared with the traditional AdaBoost algorithm, LeNet model, and MLeNet model without pre-training and fine-tuning, the method of the present invention is more suitable for the mask face detection problem by adding the MLeNet model (i.e., Ours) with pre-training and fine-tuning.
TABLE 2
| Ours | AdaBoost[2] | LeNet[11] | MLeNet | |
| Recall | 0.925 | 0.75 | 0.82 | 0.85 |
| Precision | 0.71 | 0.6 | 0.64 | 0.68 |
| F1-score | 0.803 | 0.667 | 0.719 | 0.756 |
"Ours" means MLeNet with pre-training and fine-tuning added; "MLeNet" denotes the MLeNet model without pre-training and fine-tuning.
The formula is illustrated below: (the formula variables and symbols defined can be described with reference to specific formula expressions)
Equation (1) defines the loss function of the model, which is used to measure the error of the output of the model from the original tag value.
Equation (2) is a definition of an indicative function, and is used to determine whether two values are equal, and if equal, the value is set to 1, otherwise, the value is 0.
Equation (3) defines an update rule of the stochastic gradient descent method, and the purpose of the update rule is to update variables (weight w and bias b) of each layer so that the target loss function moves along the direction of the local minimum, and a final local optimal solution is obtained.
Claims (2)
1. A robust mask face detection method based on an improved LeNet is characterized by comprising the following steps:
1) expanding training data by horizontally turning over the original training picture;
2) by modifying the structure of the traditional LeNet model, a new MLeNet model is provided, so that the MLeNet model is suitable for the detection problem of a masked human;
3) pre-training an MLeNet structure by using parameters in an original LeNet model, and finely adjusting the MLeNet model to obtain a detector suitable for a mask face;
4) and the position of the face of the masked person is accurately positioned by combining a sliding window and a non-maximization inhibition technology.
2. The robust masked face detection method based on the improved LeNet as claimed in claim 1, wherein in step 2), the specific method for proposing the new MLeNet model by modifying the structure of the traditional LeNet model to adapt to the masked human detection problem is: the size of the convolution kernel and the number of the characteristic graphs are adjusted, and in addition, the number 10 of the nodes of the original output layer is changed to be 2, so that the method is suitable for the 2-classification problem of human detection.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611127956.3A CN106778589A (en) | 2016-12-09 | 2016-12-09 | A kind of masked method for detecting human face of robust based on modified LeNet |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611127956.3A CN106778589A (en) | 2016-12-09 | 2016-12-09 | A kind of masked method for detecting human face of robust based on modified LeNet |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106778589A true CN106778589A (en) | 2017-05-31 |
Family
ID=58877630
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611127956.3A Pending CN106778589A (en) | 2016-12-09 | 2016-12-09 | A kind of masked method for detecting human face of robust based on modified LeNet |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106778589A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108921002A (en) * | 2018-04-23 | 2018-11-30 | 中国科学院自动化研究所 | Audio-video recognition methods and device are feared cruelly based on multi thread fusion |
| CN109223002A (en) * | 2018-08-27 | 2019-01-18 | 上海铱硙医疗科技有限公司 | Self-closing disease illness prediction technique, device, equipment and storage medium |
| CN115719465A (en) * | 2022-11-24 | 2023-02-28 | 北京百度网讯科技有限公司 | Vehicle detection method, apparatus, device, storage medium, and program product |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106204779A (en) * | 2016-06-30 | 2016-12-07 | 陕西师范大学 | The check class attendance method learnt based on plurality of human faces data collection strategy and the degree of depth |
-
2016
- 2016-12-09 CN CN201611127956.3A patent/CN106778589A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106204779A (en) * | 2016-06-30 | 2016-12-07 | 陕西师范大学 | The check class attendance method learnt based on plurality of human faces data collection strategy and the degree of depth |
Non-Patent Citations (1)
| Title |
|---|
| SHAOHUI LIN等: "Masked face detection via a modified LeNet", 《HTTPS://WWW.SCIENCEDIRECT.COM/SCIENCE/ARTICLE/PII/S0925231216309523》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108921002A (en) * | 2018-04-23 | 2018-11-30 | 中国科学院自动化研究所 | Audio-video recognition methods and device are feared cruelly based on multi thread fusion |
| CN109223002A (en) * | 2018-08-27 | 2019-01-18 | 上海铱硙医疗科技有限公司 | Self-closing disease illness prediction technique, device, equipment and storage medium |
| CN115719465A (en) * | 2022-11-24 | 2023-02-28 | 北京百度网讯科技有限公司 | Vehicle detection method, apparatus, device, storage medium, and program product |
| CN115719465B (en) * | 2022-11-24 | 2023-11-03 | 北京百度网讯科技有限公司 | Vehicle detection method, device, apparatus, storage medium, and program product |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20200285896A1 (en) | Method for person re-identification based on deep model with multi-loss fusion training strategy | |
| Jiang et al. | A pedestrian detection method based on genetic algorithm for optimize XGBoost training parameters | |
| Zulfiqar et al. | Deep face recognition for biometric authentication | |
| Wang et al. | Face recognition based on deep learning | |
| CN111414862B (en) | Expression recognition method based on neural network fusion key point angle change | |
| CN103605972B (en) | Non-restricted environment face verification method based on block depth neural network | |
| Li et al. | Pedestrian detection based on deep learning model | |
| CN112434599B (en) | Pedestrian re-identification method based on random occlusion recovery of noise channel | |
| CN106778589A (en) | A kind of masked method for detecting human face of robust based on modified LeNet | |
| Xu et al. | Visual-semantic transformer for face forgery detection | |
| Chen et al. | A multi-scale fusion convolutional neural network for face detection | |
| Hdioud et al. | Facial expression recognition of masked faces using deep learning | |
| Li et al. | Feature extraction based on deep‐convolutional neural network for face recognition | |
| Wang et al. | A prototype-oriented contrastive adaption network for cross-domain facial expression recognition | |
| Li | A discriminative learning convolutional neural network for facial expression recognition | |
| Nasiri et al. | Video Surveillance Framework Based on Real-Time Face Mask Detection and Recognition | |
| Liu et al. | Enhanced attacks on defensively distilled deep neural networks | |
| Song et al. | Deep convolutional neural networks with adaptive spatial feature for person re-identification | |
| Wu et al. | Masked Face Recognition Using FaceNet | |
| Qi et al. | YOLOFKP: dense face detection based on YOLOv3 key point network | |
| Zhenzhou et al. | Face recognition based on improved residual neural network | |
| Gooya et al. | Robust and discriminating face recognition system based on a neural network and correlation techniques | |
| Wang et al. | Driver Fatigue Detection Using Improved Deep Learning and Personalized Framework | |
| You et al. | Deep domain adaptation with a few samples for face identification | |
| Noh et al. | Deep-learning-based face recognition for worker access control management in hazardous areas |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170531 |