CN105760859A - Method and device for identifying reticulate pattern face image based on multi-task convolutional neural network - Google Patents
Method and device for identifying reticulate pattern face image based on multi-task convolutional neural network Download PDFInfo
- Publication number
- CN105760859A CN105760859A CN201610165633.7A CN201610165633A CN105760859A CN 105760859 A CN105760859 A CN 105760859A CN 201610165633 A CN201610165633 A CN 201610165633A CN 105760859 A CN105760859 A CN 105760859A
- Authority
- CN
- China
- Prior art keywords
- reticulate pattern
- image
- convolutional neural
- training
- neural networks
- 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.)
- Granted
Links
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 72
- 238000000034 method Methods 0.000 title claims abstract description 41
- 238000012549 training Methods 0.000 claims abstract description 67
- 230000001815 facial effect Effects 0.000 claims description 55
- 230000008569 process Effects 0.000 claims description 14
- 230000007935 neutral effect Effects 0.000 claims description 8
- 230000006870 function Effects 0.000 abstract description 19
- 230000000694 effects Effects 0.000 abstract description 9
- 230000009466 transformation Effects 0.000 abstract description 6
- 238000013461 design Methods 0.000 abstract description 2
- 238000005516 engineering process Methods 0.000 description 10
- 238000010586 diagram Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 238000011084 recovery Methods 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
- G06V40/172—Classification, e.g. identification
-
- 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
Abstract
The present invention discloses a method and a device for identifying a reticulate pattern face image based on a multi-task convolutional neural network. The method comprises the steps of: collecting reticulate pattern face image and corresponding clear face image pairs, then using the multi-task convolutional neural network to respectively design object functions based on regression and classification, training a face image reticulate pattern removing model, and finally inputting the reticulate pattern face image into the trained reticulate pattern removing model to obtain a face image without reticulate pattern, thereby performing subsequent face image identification tasks. According to the method, a multi-task learning frame is adopted, the task for restoring a reticulate pattern image to a clear image is expressed as two object functions which are assistant with each other, and the convolutional neural network is utilized to learn complicated nonlinear transformation referred therein. The method not only effectively improves convergence rate during model training, but also can greatly improve image restoration effect and generalization ability, thereby greatly improving identification accuracy rate of the reticulate pattern face image.
Description
Technical field
The present invention relates to computer vision, pattern recognition, the technical field such as machine learning, particularly a kind of reticulate pattern facial image recognition method based on multitask convolutional neural networks (Multi-taskConvNetforFaceRecognition is called for short MTCN).
Background technology
As the one of biometrics identification technology, recognition of face, due to its feature untouchable and accurate, convenient, has good development and application prospect.Face recognition technology has all played highly important effect, such as airport security, frontier inspection clearance etc. in many application scenarios.Traditional face recognition technology mainly under Same Scene, the data gathered in different periods.But identify the raising of the level of technology along with current face, in order to use face recognition technology more easily.The face recognition technology of identity-based card-living photo comparison starts to obtain more concern gradually.This technology can be thrown in more easily in airport clearance and bank such as remotely open an account at the application scenarios, it is possible to need not register in advance and just can correctly identify, convenience when face recognition technology use is greatly improved.
Current identity card-living photo comparison technology or the difficult problem of a comparison.On the one hand, owing to gathering the difference of environment and collecting device, substantially, this is a heterogeneous recognition of face problem.In heterogeneous recognition of face problem, owing in the class that numerous reasons cause, difference is very big, thus it is very difficult to solve.On the other hand, in order to protect the ID Card Image of user not abused by lawless person, the identity card picture read out by card reader of ID card is mostly added random reticulate pattern and does watermark.It is protected that these moire make privacy of user be able to, and but machine is carried out recognition of face and causes huge interference.Random reticulate pattern greatly reduces the accuracy rate of the steps such as Face datection, critical point detection and feature extraction, thus having a strong impact on final recognition effect.
In recent years, degree of depth study all achieves, in the various fields of machine vision, the effect attracted people's attention.The model wherein attracted attention the most surely belongs to convolutional neural networks, and this class model uses multilamellar convolutional layer and pond layer, it is possible to extract hierarchical feature, it is achieved stronger non-linear expression.Convolutional neural networks, in fields such as object classification, action recognition, image segmentation and recognitions of face, all achieves the effect being significantly stronger than traditional method.In some Low Level Vision problems, such as image denoising, Image Super-resolution, in the problem such as image deblurring, this model also all achieves good effect.
In order to solve the identification problem of reticulate pattern facial image, we can pass through algorithm and recover the clear face image without reticulate pattern from reticulate pattern facial image.Utilize convolutional neural networks as above can train a model for directly recovering facial image clearly from reticulate pattern human face data.But simply using reticulate pattern image as input, picture rich in detail is as output, and training neutral net effectiveness comparison out is general, and trains process veryer long, restrains slower.It is additionally, since training process and lacks certain priori guiding, the model generalization poor-performing of training.
Summary of the invention
Accuracy rate during in order to improve reticulate pattern facial image for recognition of face, the present invention proposes a kind of reticulate pattern facial image recognition method based on multitask convolutional neural networks.Accuracy rate during in order to improve reticulate pattern image for recognition of face, first the present invention recovers the clear face image without reticulate pattern with reticulate pattern image.The present invention uses model framework based on convolutional neural networks, priori according to specific tasks, devise the Optimized model of a multitask, optimize residual error to return and reticulate pattern two tasks of prediction simultaneously, optimize the convergence rate being conducive to promoting network training simultaneously, and final model has convergence rate and recovery effects faster, and be effectively increased the generalization ability of model.
The present invention proposes a kind of reticulate pattern facial image recognition method based on multitask convolutional neural networks, specifically implements according to following steps:
The clear face image of step S1, collection reticulate pattern facial image and correspondence is to as sample image pair, forming training dataset, for each sample image to obtaining indicating the label figure of reticulate pattern position by threshold method.
Step S2, training obtain going out the convolutional neural networks model of the clear face image without reticulate pattern from reticulate pattern face image restoration, including:
Utilize the sample image pair that described training data is concentrated, train a multitask convolutional neural networks, in training process, the reticulate pattern facial image of input is processed by multiple convolutional layers of this multitask convolutional neural networks first half, latter half is divided into two tasks, is utilized respectively the data after process and corresponding loss object function is trained；Main task in wherein said two tasks, for returning clear face image and the difference of reticulate pattern facial image, obtains residual image；Nonproductive task, for predicting the reticulate pattern position of reticulate pattern facial image, obtains the reticulate pattern image of prediction；The convolutional neural networks model finally trained be finally output as described residual image and the addition of reticulate pattern position, namely without the clear face image of reticulate pattern；
The convolutional neural networks model that step S3, use train, recovers clear face image to be identified, and uses clear face image to be identified to carry out recognition of face.
The invention allows for a kind of reticulate pattern facial image identification device based on multitask convolutional neural networks, including:
Training sample acquisition module, for collecting reticulate pattern facial image and corresponding clear face image to as sample image pair, forming training dataset, for each sample image to obtaining indicating the label figure of reticulate pattern position by threshold method.
Convolutional neural networks training module, for training the convolutional neural networks model obtaining going out the clear face image without reticulate pattern from reticulate pattern face image restoration, including:
Utilize the sample image pair that described training data is concentrated, train a multitask convolutional neural networks, in training process, the reticulate pattern facial image of input is processed by multiple convolutional layers of this multitask convolutional neural networks first half, latter half is divided into two tasks, is utilized respectively the data after process and corresponding loss object function is trained；Main task in wherein said two tasks, for returning clear face image and the difference of reticulate pattern facial image, obtains residual image；Nonproductive task, for predicting the reticulate pattern position of reticulate pattern facial image, obtains the reticulate pattern image of prediction；The image recognition model finally trained be finally output as described residual image and the addition of reticulate pattern position, namely without the clear face image of reticulate pattern；
Identification module, for using the convolutional neural networks model trained, recovers clear face image to be identified, and uses clear face image to be identified to carry out recognition of face.
Beneficial effects of the present invention: the present invention is directed to a particular problem in recognition of face, identity card photograph-living photo comparison problem a, it is proposed that method being widely applied meaning.The method can recover the clear face image without reticulate pattern by multitask convolutional neural networks from reticulate pattern image, and utilizes the clear face image recovered to carry out recognition of face.The multitask convolutional neural networks model that the present invention proposes, employs the multiobject optimal way of multitask so that model is restrained faster, better effects if, and Generalization Capability is higher.By the first descreening identifying schemes again that the present invention proposes, it is possible to increase substantially the accuracy rate of the recognition of face of reticulate pattern image.
Accompanying drawing explanation
Fig. 1 is the example of the reticulate pattern facial image studied of the present invention and clear face image；
Fig. 2 is the method flow diagram of the reticulate pattern facial image recognition method in the present invention based on multitask convolutional neural networks；
Fig. 3 is the network diagram of multitask convolutional neural networks model in the present invention.
Detailed description of the invention
For making the object, technical solutions and advantages of the present invention clearly understand, below in conjunction with specific embodiment, and with reference to accompanying drawing, the present invention is described in further detail.
The present invention, by the conversion of convolutional neural networks one nonlinearity of study of multitask, is used for recovering the picture rich in detail without reticulate pattern from reticulate pattern image, and uses picture rich in detail to carry out follow-up recognition of face.
Fig. 1 is the exemplary diagram of reticulate pattern facial image used in the present invention and clear face image.
Fig. 2 is a kind of reticulate pattern facial image recognition method flow chart based on multitask convolutional neural networks that the present invention proposes, and the method includes following step as shown in Figure 2:
Step S1, gathers reticulate pattern facial image and corresponding clear face image to as training dataset, described image pair in the same size, gray-scale map or coloured image；To described image to doing difference, obtain the bianry image of instruction reticulate pattern distributing position as label figure according to certain threshold value.Wherein, reticulate pattern facial image is expressed as x, and corresponding clear face figure is expressed as y, and corresponding label figure is expressed as lr.
Step S2, utilizes the image pair that described training data is concentrated, and trains multitask convolutional neural networks model, to be used for recovering facial image y clearly from reticulate pattern facial image x.Identical in order to ensure input and output image size, convolutional neural networks only comprises convolutional layer, and all adds the operation of benefit limit at each convolutional layer, namely by expanded images edge pixel so that after convolution operation, the size of image is constant.In one embodiment, described convolutional neural networks structure is made up of 6 layers of convolutional layer and corresponding correction linear unit, and wherein every layer of convolutional layer has the wave filter of 64 3*3 sizes.The number of the number of plies of convolutional layer and every layer of convolutional layer median filter and big I carry out selecting to arrange according to practical situation.As it is shown on figure 3, by the input layer in original image input to described convolutional neural networks, after multilamellar convolutional layer and nonlinear transformation layer operation, obtain convolution results and nonlinear transformation result；In task one, the convolution of abovementioned layers and nonlinear transformation result are input in the wave filter of an only one of which convolution kernel, obtaining a size characteristic pattern identical with input picture, this, as the convolutional neural networks prediction to residual error, is used for calculating the loss of task one together with actual value.In task two, equally convolution and the nonlinear transformation result of abovementioned layers are input in the wave filter of an other only one of which convolution kernel, and the characteristic pattern obtained is carried out nonlinear transformation through a S type function, the result of conversion as the result polluting position prediction, can calculate target loss together with the actual value provided in training set.
In this step, utilize the sample image pair that described training data is concentrated, train a multitask convolutional neural networks, in training process, the reticulate pattern facial image of input is processed by multiple convolutional layers of this multitask convolutional neural networks first half, latter half is divided into two tasks, is utilized respectively the data after process and corresponding loss object function is trained；Main task in wherein said two tasks, for returning clear face image and the difference of reticulate pattern facial image, obtains residual image；Nonproductive task, for predicting the reticulate pattern position of reticulate pattern facial image, obtains the reticulate pattern image of prediction；The convolutional neural networks model finally trained be finally output as described residual image and the addition of reticulate pattern position, namely without the clear face image of reticulate pattern.
The present invention utilizes the nonlinearity capability of fitting of described convolutional neural networks, these specific tasks of picture rich in detail are recovered for reticulate pattern image, constructing with reticulate pattern image for input, the error image of picture rich in detail and reticulate pattern image is as the convolutional neural networks model of output.Especially, this network can return out picture rich in detail and the difference of reticulate pattern image, rather than directly returns out picture rich in detail.So, by network as shown in Figure 2, use two relevant tasks, it is possible to train a convolutional neural networks that can recover picture rich in detail, such that it is able to reticulate pattern face image restoration to be become the clear face image not having reticulate pattern to pollute.Although the output of network is the difference of picture rich in detail and reticulate pattern image, but when identifying, it is only necessary to the picture rich in detail that the output of network and reticulate pattern image addition can be restored out.
Concrete, two tasks of multiple target convolutional neural networks respectively return the difference of picture rich in detail and reticulate pattern image and the position of prediction reticulate pattern.Especially, return difference object function can by minimize with mean square error be representative a series of object functions complete, such as:
Wherein, J1(w1) represent that picture rich in detail returns the object function of difference, w1For the parameter of the needs training that this object function relates to, r=x-y is reticulate pattern image x and the residual error of prediction picture rich in detail y,Representing the picture rich in detail recurrence device that neutral net is acquired, i, j represents the pixel coordinate in image, and N represents the training sample sum that training data is concentrated.
The task of prediction reticulate pattern position can be expressed as whether a certain pixel of predicted picture is this two classification problem of reticulate pattern.This problem can with minimize using logistical regression loss be representative a series of Classification Loss functions as optimization aim, such as:
Wherein, J2(w2) represent the object function predicting reticulate pattern position, w2Parameter for the needs training that this object function relates to, φ represents and represents the reticulate pattern position forecaster that neutral net is acquired, l represents the bianry image of reticulate pattern position in input picture, its size is the same with input picture, it is used for characterizing which region of input picture contaminated, wherein pixel is the region of zero is uncontaminated region, and pixel is 1 and illustrates that this region is contaminated.
Above-mentioned convolutional neural networks, the task of regression residuals is as main task, and predicts that the task of reticulate pattern position is nonproductive task, and the final goal function of described convolutional neural networks is: J (w1, w2)=J1(w1)+αJ2(w2), α is the weight parameter of nonproductive task.In above-mentioned object function, the parameter in all parameters and abovementioned layers carries out such as through minimizing above-mentioned total object function.The training of model can be undertaken by back-propagation algorithm, and continuous iteration updates each layer parameter, minimizes this object function.
Described convolutional neural networks is trained as follows:
Two tasks are added identical weight, namely initialize α=1 by step S21: the initial stage of network training；
Step S22: the sample input concentrated by described training data is trained as described convolutional neural networks, until total object function J (w1, w2)=J1(w1)+αJ2(w2) tend towards stability；
Step S23: reduce the weight parameter of described nonproductive task, goes to step S22 and continues training, until the weight parameter of described nonproductive task reduces to 0；
Step S24: continue training, until training loss J (w1, w2)=J1(w1)+αJ2(w2) do not continue to reduce, at this moment can preserve the currency of each layer parameter in network, as the parameter of final mask.
Step S3, for new reticulate pattern facial image, is entered in the convolutional neural networks trained, obtains the difference of picture rich in detail and reticulate pattern image, and this value and reticulate pattern image addition can recover the clear face image without reticulate pattern.Next can by traditional recognition of face step, through Face datection, critical point detection and after feature extraction, carry out corresponding aspect ratio pair, complete recognition of face task.
In order to describe the specific embodiment of the present invention and checking effectiveness of the invention in detail, the recognition of face task of method one reticulate pattern image of application that the present invention is proposed by we.Concrete, in order to train the multitask convolutional neural networks model of descreening, we prepared 500,000 facial images with reticulate pattern and its correspondence without reticulate pattern picture rich in detail, and calculate the label figure of the reticulate pattern position of instruction correspondence image pair.Utilize network structure and object function that we design, with reticulate pattern facial image for input, utilize gradient anti-pass to train this neutral net.Training process constantly adjusts the weight of different task, until last network convergence, obtains the model for recovering clear face image.
In order to test the effectiveness of this model, we have additionally prepared the identity card picture clearly (everyone) of 300 people and corresponding individual living photo 300, it is notable that these 300 people are not present in training set.This data set can be used to test the effect of algorithm during identity card-living photo comparison.Using clear identity card according to when comparing with living photo, accuracy rate when using the depth characteristic of the heterogeneous recognition of face of a kind of special disposal to be identified is such as shown in table 1 first row.Follow-up experiment, we use this feature to carry out recognition of face.Discrimination time for calibration tape reticulate pattern image for recognition of face, we give these 300 the random reticulate patterns of face addition clearly, generate the identity card with reticulate pattern and shine.This series of images and corresponding clear living photo is used to carry out recognition of face, it has been found that its discrimination (such as table 1) is greatly lowered.Afterwards, we use the multitask convolutional neural networks model trained, and from above-mentioned with recovering facial image clearly the facial image of reticulate pattern, then carry out recognition of face with corresponding living photo, and concrete recognition result is as shown in table 1.Although also having certain gap with the discrimination of original clear picture, but have compared with the recognition result with reticulate pattern facial image and significantly promoted.These embodiment valid certificates method proposed by the invention effectiveness to reticulate pattern facial image identification.
Table 1 is the accuracy rate comparing result table that the face recognition accuracy rate after the present invention processes is for recognition of face with untreated reticulate pattern image and normal picture rich in detail, as follows:
| TPRFPR=1% | TPRFPR=0.1% | TPRFPR=0.01% | |
| Picture rich in detail | 91.20 | 75.20 | 50.20 |
| Reticulate pattern image | 49.00 | 30.80 | 16.40 |
| Descreening image | 81.00 | 57.80 | 29.20 |
Particular embodiments described above; the purpose of the present invention, technical scheme and beneficial effect have been further described; it it should be understood that; the foregoing is only specific embodiments of the invention; it is not limited to the present invention; all within the spirit and principles in the present invention, any amendment of making, equivalent replacement, improvement etc., should be included within protection scope of the present invention.
Claims (10)
1. the reticulate pattern facial image recognition method based on multitask convolutional neural networks, it is characterised in that specifically implement according to following steps:
The clear face image of step S1, collection reticulate pattern facial image and correspondence is to as sample image pair, forming training dataset, for each sample image to obtaining indicating the label figure of reticulate pattern position by threshold method.
Step S2, training obtain going out the convolutional neural networks model of the clear face image without reticulate pattern from reticulate pattern face image restoration, including:
Utilize the sample image pair that described training data is concentrated, train a multitask convolutional neural networks, in training process, the reticulate pattern facial image of input is processed by multiple convolutional layers of this multitask convolutional neural networks first half, latter half is divided into two tasks, is utilized respectively the data after process and corresponding loss object function is trained；Main task in wherein said two tasks, for returning clear face image and the difference of reticulate pattern facial image, obtains residual image；Nonproductive task, for predicting the reticulate pattern position of reticulate pattern facial image, obtains the reticulate pattern image of prediction；The convolutional neural networks model finally trained be finally output as described residual image and the addition of reticulate pattern position, namely without the clear face image of reticulate pattern；
The convolutional neural networks model that step S3, use train, recovers clear face image to be identified, and uses clear face image to be identified to carry out recognition of face.
2. the reticulate pattern facial image recognition method based on multitask convolutional neural networks according to claim 1, it is characterized in that, in described step S1, the reticulate pattern facial image gathered is in the same size with corresponding clear face image pair, and each sample image concentrated for training data obtains the bianry image indicating the position of reticulate pattern distribution as label figure to according to certain threshold value.
3. the reticulate pattern facial image recognition method based on multitask convolutional neural networks according to claim 1, it is characterised in that described step S2 includes:
Step S21: initializing the weight parameter making two tasks equal, wherein, total object function is J (w1, w2)=J1(w1)+αJ2(w2), J1(w1) represent main task object function, J2(w2) for the object function of nonproductive task；w1、w2The respectively training parameter in main task and nonproductive task object function；α is the weight parameter of nonproductive task, initializes α=1 in this step；
Step S22: the sample input concentrated by described training data is trained as described convolutional neural networks, until total object function J (w1, w2)=J1(w1)+αJ2(w2) tend towards stability, wherein；
Step S23: reduce the weight parameter of described nonproductive task, goes to step S22 and continues training, until the weight parameter of described nonproductive task reduces to 0；
Step S24: continue training, until training loss no longer reduces, thus obtaining final convolutional neural networks model.
4. the reticulate pattern facial image recognition method based on multitask convolutional neural networks as claimed in claim 3, it is characterised in that the object function of main task is expressed as:
Wherein, r=x-y is reticulate pattern image x and the residual error of prediction picture rich in detail y,Representing the picture rich in detail recurrence device that neutral net is acquired, i, j represents the pixel coordinate in image, and N represents the training sample sum that training data is concentrated.
5. the reticulate pattern facial image recognition method based on multitask convolutional neural networks as claimed in claim 3, it is characterised in that the object function of nonproductive task is expressed as:
6. the reticulate pattern facial image identification device based on multitask convolutional neural networks, it is characterised in that including:
Training sample acquisition module, for collecting reticulate pattern facial image and corresponding clear face image to as sample image pair, forming training dataset, for each sample image to obtaining indicating the label figure of reticulate pattern position by threshold method.
Convolutional neural networks training module, for training the convolutional neural networks model obtaining going out the clear face image without reticulate pattern from reticulate pattern face image restoration, including:
Utilize the sample image pair that described training data is concentrated, train a multitask convolutional neural networks, in training process, the reticulate pattern facial image of input is processed by multiple convolutional layers of this multitask convolutional neural networks first half, latter half is divided into two tasks, is utilized respectively the data after process and corresponding loss object function is trained；Main task in wherein said two tasks, for returning clear face image and the difference of reticulate pattern facial image, obtains residual image；Nonproductive task, for predicting the reticulate pattern position of reticulate pattern facial image, obtains the reticulate pattern image of prediction；The image recognition model finally trained be finally output as described residual image and the addition of reticulate pattern position, namely without the clear face image of reticulate pattern；
Identification module, for using the convolutional neural networks model trained, recovers clear face image to be identified, and uses clear face image to be identified to carry out recognition of face.
7. the reticulate pattern facial image identification device based on multitask convolutional neural networks according to claim 6, it is characterized in that, the reticulate pattern facial image gathered is in the same size with corresponding clear face image pair, and each sample image concentrated for training data obtains the bianry image indicating the position of reticulate pattern distribution as label figure to according to certain threshold value.
8. the reticulate pattern facial image identification device based on multitask convolutional neural networks according to claim 6, it is characterised in that described convolutional neural networks module includes:
Initialization module, for initializing so that the weight parameter of two tasks is equal, wherein, total object function is J (w1, w2)=J1(w1)+αJ2(w2), J1(w1) represent main task object function, J2(w2) for the object function of nonproductive task；w1、w2The respectively training parameter in main task and nonproductive task object function；α is the weight parameter of nonproductive task, initializes α=1 in this step；
Initial training module, is trained as described convolutional neural networks for the sample input concentrated by described training data, until total object function J (w1, w2)=J1(w1)+αJ2(w2) tend towards stability, wherein；
Degree of depth training module, for reducing the weight parameter of described nonproductive task, turns initial training module and continues training, until the weight parameter of described nonproductive task reduces to 0；
Training result output module, is used for continuing training, until training loss no longer reduces, thus obtaining final image recognition model.
9. the reticulate pattern facial image identification device based on multitask convolutional neural networks as claimed in claim 8, it is characterised in that the object function of main task is expressed as:
Wherein, r=x-y is reticulate pattern image x and the residual error of prediction picture rich in detail y,Representing the picture rich in detail recurrence device that neutral net is acquired, i, j represents the pixel coordinate in image, and N represents the training sample sum that training data is concentrated.
10. the reticulate pattern facial image identification device based on multitask convolutional neural networks as claimed in claim 8, it is characterised in that the object function of nonproductive task is expressed as:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610165633.7A CN105760859B (en) | 2016-03-22 | 2016-03-22 | Reticulate pattern facial image recognition method and device based on multitask convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610165633.7A CN105760859B (en) | 2016-03-22 | 2016-03-22 | Reticulate pattern facial image recognition method and device based on multitask convolutional neural networks |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105760859A true CN105760859A (en) | 2016-07-13 |
| CN105760859B CN105760859B (en) | 2018-12-21 |
Family
ID=56345590
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610165633.7A Active CN105760859B (en) | 2016-03-22 | 2016-03-22 | Reticulate pattern facial image recognition method and device based on multitask convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105760859B (en) |
Cited By (35)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106295678A (en) * | 2016-07-27 | 2017-01-04 | 北京旷视科技有限公司 | Neural metwork training and construction method and device and object detection method and device |
| CN106529402A (en) * | 2016-09-27 | 2017-03-22 | 中国科学院自动化研究所 | Multi-task learning convolutional neural network-based face attribute analysis method |
| CN106548159A (en) * | 2016-11-08 | 2017-03-29 | 中国科学院自动化研究所 | Reticulate pattern facial image recognition method and device based on full convolutional neural networks |
| CN106815566A (en) * | 2016-12-29 | 2017-06-09 | 天津中科智能识别产业技术研究院有限公司 | A kind of face retrieval method based on multitask convolutional neural networks |
| CN107491767A (en) * | 2017-08-31 | 2017-12-19 | 广州云从信息科技有限公司 | End to end without constraint face critical point detection method |
| CN107545277A (en) * | 2017-08-11 | 2018-01-05 | 腾讯科技（上海）有限公司 | Model training, auth method, device, storage medium and computer equipment |
| CN107563372A (en) * | 2017-07-20 | 2018-01-09 | 济南中维世纪科技有限公司 | A kind of license plate locating method based on deep learning SSD frameworks |
| CN107766844A (en) * | 2017-11-13 | 2018-03-06 | 杭州有盾网络科技有限公司 | Method, apparatus, equipment of a kind of reticulate pattern according to recognition of face |
| CN107798667A (en) * | 2017-11-23 | 2018-03-13 | 中电科新型智慧城市研究院有限公司 | Face Enhancement Method based on residual error study |
| CN107993190A (en) * | 2017-11-14 | 2018-05-04 | 中国科学院自动化研究所 | Image watermark removal device |
| CN107992783A (en) * | 2016-10-26 | 2018-05-04 | 上海银晨智能识别科技有限公司 | Face image processing process and device |
| CN108205701A (en) * | 2016-12-20 | 2018-06-26 | 联发科技股份有限公司 | A kind of system and method for performing convolutional calculation |
| CN108205802A (en) * | 2016-12-23 | 2018-06-26 | 北京市商汤科技开发有限公司 | Deep neural network model training, image processing method and device and equipment |
| CN108229349A (en) * | 2017-12-21 | 2018-06-29 | 中国科学院自动化研究所 | Reticulate pattern facial image identification device |
| CN108399649A (en) * | 2018-03-05 | 2018-08-14 | 中科视拓（北京）科技有限公司 | A kind of single picture three-dimensional facial reconstruction method based on cascade Recurrent networks |
| CN108510454A (en) * | 2018-03-21 | 2018-09-07 | 百度在线网络技术（北京）有限公司 | Method and apparatus for generating depth image |
| CN108596818A (en) * | 2017-12-22 | 2018-09-28 | 华南理工大学 | A kind of image latent writing analysis method based on multi-task learning convolutional neural networks |
| CN108734673A (en) * | 2018-04-20 | 2018-11-02 | 平安科技（深圳）有限公司 | Descreening systematic training method, descreening method, apparatus, equipment and medium |
| CN109002790A (en) * | 2018-07-11 | 2018-12-14 | 广州视源电子科技股份有限公司 | A kind of method, apparatus of recognition of face, equipment and storage medium |
| CN109147940A (en) * | 2018-07-05 | 2019-01-04 | 北京昆仑医云科技有限公司 | From the device and system of the medical image automatic Prediction physiological status of patient |
| CN109543763A (en) * | 2018-11-28 | 2019-03-29 | 重庆大学 | A kind of Raman spectrum analysis method based on convolutional neural networks |
| CN109801225A (en) * | 2018-12-06 | 2019-05-24 | 重庆邮电大学 | Face reticulate pattern stain minimizing technology based on the full convolutional neural networks of multitask |
| CN110008817A (en) * | 2019-01-29 | 2019-07-12 | 北京奇艺世纪科技有限公司 | Model training, image processing method, device, electronic equipment and computer readable storage medium |
| CN110070027A (en) * | 2019-04-17 | 2019-07-30 | 南京邮电大学 | Pedestrian based on intelligent internet of things system recognition methods again |
| CN110084143A (en) * | 2019-04-04 | 2019-08-02 | 广州大学 | A kind of emotional information guard method and system for recognition of face |
| CN110084147A (en) * | 2019-04-04 | 2019-08-02 | 广州大学 | A kind of gender method for secret protection and system for recognition of face |
| WO2019165006A1 (en) * | 2018-02-22 | 2019-08-29 | Mcafee, Llc | Image hidden information detector |
| CN110232352A (en) * | 2019-06-12 | 2019-09-13 | 东北大学 | A kind of improved method of the multitask concatenated convolutional neural network model for recognition of face |
| CN110263756A (en) * | 2019-06-28 | 2019-09-20 | 东北大学 | A kind of human face super-resolution reconstructing system based on joint multi-task learning |
| CN110647805A (en) * | 2019-08-09 | 2020-01-03 | 平安科技（深圳）有限公司 | Reticulated image identification method and device and terminal equipment |
| CN111260548A (en) * | 2018-11-30 | 2020-06-09 | 浙江宇视科技有限公司 | Mapping method and device based on deep learning |
| CN111612133A (en) * | 2020-05-20 | 2020-09-01 | 广州华见智能科技有限公司 | Internal organ feature coding method based on face image multi-stage relation learning |
| US20210141882A1 (en) * | 2019-11-12 | 2021-05-13 | Electronics And Telecommunications Research Institute | Apparatus and method for authenticating user based on multiple biometric information |
| US11341631B2 (en) | 2017-08-09 | 2022-05-24 | Shenzhen Keya Medical Technology Corporation | System and method for automatically detecting a physiological condition from a medical image of a patient |
| US11423634B2 (en) | 2018-08-03 | 2022-08-23 | Huawei Cloud Computing Technologies Co., Ltd. | Object detection model training method, apparatus, and device |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090116747A1 (en) * | 2006-08-01 | 2009-05-07 | Duong Tuan A | Artificial intelligence systems for identifying objects |
| CN105184303A (en) * | 2015-04-23 | 2015-12-23 | 南京邮电大学 | Image marking method based on multi-mode deep learning |
-
2016
- 2016-03-22 CN CN201610165633.7A patent/CN105760859B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090116747A1 (en) * | 2006-08-01 | 2009-05-07 | Duong Tuan A | Artificial intelligence systems for identifying objects |
| CN105184303A (en) * | 2015-04-23 | 2015-12-23 | 南京邮电大学 | Image marking method based on multi-mode deep learning |
Non-Patent Citations (4)
| Title |
|---|
| ABRAR H. ABDULNABI等: "Multi-Task CNN Model for Attribute Prediction", 《IEEE TRANSACTIONS ON MULTIMEDIA》 * |
| DONG CAO等: "Joint space learning for video-based face recognition", 《2015 3RD IAPR ASIAN CONFERENCE ON PATTERN RECOGNITION (ACPR)》 * |
| GUOSHENG HU等: "When Face Recognition Meets with Deep Learning: An Evaluation of Convolutional Neural Networks for Face Recognition", 《2015 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION WORKSHOP (ICCVW)》 * |
| YIZHANG XIA等: "Face occlusion detection based on multi-task convolution neural network", 《2015 12TH INTERNATIONAL CONFERENCE ON FUZZY SYSTEMS AND KNOWLEDGE DISCOVERY (FSKD)》 * |
Cited By (59)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106295678A (en) * | 2016-07-27 | 2017-01-04 | 北京旷视科技有限公司 | Neural metwork training and construction method and device and object detection method and device |
| US10769493B2 (en) | 2016-07-27 | 2020-09-08 | Beijing Kuangshi Technology Co., Ltd. | Method and apparatus for neural network training and construction and method and apparatus for object detection |
| CN106529402A (en) * | 2016-09-27 | 2017-03-22 | 中国科学院自动化研究所 | Multi-task learning convolutional neural network-based face attribute analysis method |
| CN106529402B (en) * | 2016-09-27 | 2019-05-28 | 中国科学院自动化研究所 | The face character analysis method of convolutional neural networks based on multi-task learning |
| CN107992783A (en) * | 2016-10-26 | 2018-05-04 | 上海银晨智能识别科技有限公司 | Face image processing process and device |
| CN106548159A (en) * | 2016-11-08 | 2017-03-29 | 中国科学院自动化研究所 | Reticulate pattern facial image recognition method and device based on full convolutional neural networks |
| CN108205701A (en) * | 2016-12-20 | 2018-06-26 | 联发科技股份有限公司 | A kind of system and method for performing convolutional calculation |
| CN108205701B (en) * | 2016-12-20 | 2021-12-28 | 联发科技股份有限公司 | System and method for executing convolution calculation |
| CN108205802A (en) * | 2016-12-23 | 2018-06-26 | 北京市商汤科技开发有限公司 | Deep neural network model training, image processing method and device and equipment |
| CN106815566A (en) * | 2016-12-29 | 2017-06-09 | 天津中科智能识别产业技术研究院有限公司 | A kind of face retrieval method based on multitask convolutional neural networks |
| CN106815566B (en) * | 2016-12-29 | 2021-04-16 | 天津中科智能识别产业技术研究院有限公司 | Face retrieval method based on multitask convolutional neural network |
| CN107563372A (en) * | 2017-07-20 | 2018-01-09 | 济南中维世纪科技有限公司 | A kind of license plate locating method based on deep learning SSD frameworks |
| CN107563372B (en) * | 2017-07-20 | 2021-01-29 | 济南中维世纪科技有限公司 | License plate positioning method based on deep learning SSD frame |
| US11341631B2 (en) | 2017-08-09 | 2022-05-24 | Shenzhen Keya Medical Technology Corporation | System and method for automatically detecting a physiological condition from a medical image of a patient |
| CN107545277A (en) * | 2017-08-11 | 2018-01-05 | 腾讯科技（上海）有限公司 | Model training, auth method, device, storage medium and computer equipment |
| CN107545277B (en) * | 2017-08-11 | 2023-07-11 | 腾讯科技（上海）有限公司 | Model training, identity verification method and device, storage medium and computer equipment |
| CN107491767B (en) * | 2017-08-31 | 2020-06-12 | 云从科技集团股份有限公司 | End-to-end unconstrained face key point detection method |
| CN107491767A (en) * | 2017-08-31 | 2017-12-19 | 广州云从信息科技有限公司 | End to end without constraint face critical point detection method |
| CN107766844A (en) * | 2017-11-13 | 2018-03-06 | 杭州有盾网络科技有限公司 | Method, apparatus, equipment of a kind of reticulate pattern according to recognition of face |
| CN107993190B (en) * | 2017-11-14 | 2020-05-19 | 中国科学院自动化研究所 | Image watermark removing device |
| CN107993190A (en) * | 2017-11-14 | 2018-05-04 | 中国科学院自动化研究所 | Image watermark removal device |
| CN107798667A (en) * | 2017-11-23 | 2018-03-13 | 中电科新型智慧城市研究院有限公司 | Face Enhancement Method based on residual error study |
| CN108229349B (en) * | 2017-12-21 | 2020-09-01 | 中国科学院自动化研究所 | Reticulate pattern human face image recognition device |
| CN108229349A (en) * | 2017-12-21 | 2018-06-29 | 中国科学院自动化研究所 | Reticulate pattern facial image identification device |
| CN108596818B (en) * | 2017-12-22 | 2021-10-26 | 华南理工大学 | Image steganalysis method based on multitask learning convolutional neural network |
| CN108596818A (en) * | 2017-12-22 | 2018-09-28 | 华南理工大学 | A kind of image latent writing analysis method based on multi-task learning convolutional neural networks |
| CN112088378A (en) * | 2018-02-22 | 2020-12-15 | 迈克菲有限责任公司 | Image hidden information detector |
| WO2019165006A1 (en) * | 2018-02-22 | 2019-08-29 | Mcafee, Llc | Image hidden information detector |
| US10699358B2 (en) | 2018-02-22 | 2020-06-30 | Mcafee, Llc | Image hidden information detector |
| CN108399649A (en) * | 2018-03-05 | 2018-08-14 | 中科视拓（北京）科技有限公司 | A kind of single picture three-dimensional facial reconstruction method based on cascade Recurrent networks |
| CN108510454A (en) * | 2018-03-21 | 2018-09-07 | 百度在线网络技术（北京）有限公司 | Method and apparatus for generating depth image |
| WO2019200702A1 (en) * | 2018-04-20 | 2019-10-24 | 平安科技（深圳）有限公司 | Descreening system training method and apparatus, descreening method and apparatus, device, and medium |
| CN108734673A (en) * | 2018-04-20 | 2018-11-02 | 平安科技（深圳）有限公司 | Descreening systematic training method, descreening method, apparatus, equipment and medium |
| CN109147940A (en) * | 2018-07-05 | 2019-01-04 | 北京昆仑医云科技有限公司 | From the device and system of the medical image automatic Prediction physiological status of patient |
| CN109002790A (en) * | 2018-07-11 | 2018-12-14 | 广州视源电子科技股份有限公司 | A kind of method, apparatus of recognition of face, equipment and storage medium |
| US11423634B2 (en) | 2018-08-03 | 2022-08-23 | Huawei Cloud Computing Technologies Co., Ltd. | Object detection model training method, apparatus, and device |
| US11605211B2 (en) | 2018-08-03 | 2023-03-14 | Huawei Cloud Computing Technologies Co., Ltd. | Object detection model training method and apparatus, and device |
| CN109543763B (en) * | 2018-11-28 | 2022-10-21 | 重庆大学 | Raman spectrum analysis method based on convolutional neural network |
| CN109543763A (en) * | 2018-11-28 | 2019-03-29 | 重庆大学 | A kind of Raman spectrum analysis method based on convolutional neural networks |
| CN111260548A (en) * | 2018-11-30 | 2020-06-09 | 浙江宇视科技有限公司 | Mapping method and device based on deep learning |
| CN111260548B (en) * | 2018-11-30 | 2023-07-21 | 浙江宇视科技有限公司 | Mapping method and device based on deep learning |
| CN109801225B (en) * | 2018-12-06 | 2022-12-27 | 重庆邮电大学 | Human face reticulate pattern stain removing method based on multitask full convolution neural network |
| CN109801225A (en) * | 2018-12-06 | 2019-05-24 | 重庆邮电大学 | Face reticulate pattern stain minimizing technology based on the full convolutional neural networks of multitask |
| CN110008817A (en) * | 2019-01-29 | 2019-07-12 | 北京奇艺世纪科技有限公司 | Model training, image processing method, device, electronic equipment and computer readable storage medium |
| CN110008817B (en) * | 2019-01-29 | 2021-12-28 | 北京奇艺世纪科技有限公司 | Model training method, image processing method, device, electronic equipment and computer readable storage medium |
| CN110084147B (en) * | 2019-04-04 | 2021-09-28 | 广州大学 | Gender privacy protection method and system for face recognition |
| CN110084147A (en) * | 2019-04-04 | 2019-08-02 | 广州大学 | A kind of gender method for secret protection and system for recognition of face |
| CN110084143B (en) * | 2019-04-04 | 2021-09-28 | 广州大学 | Emotion information protection method and system for face recognition |
| CN110084143A (en) * | 2019-04-04 | 2019-08-02 | 广州大学 | A kind of emotional information guard method and system for recognition of face |
| CN110070027A (en) * | 2019-04-17 | 2019-07-30 | 南京邮电大学 | Pedestrian based on intelligent internet of things system recognition methods again |
| CN110232352B (en) * | 2019-06-12 | 2022-11-25 | 东北大学 | Improvement method of multi-task cascade convolution neural network model for face recognition |
| CN110232352A (en) * | 2019-06-12 | 2019-09-13 | 东北大学 | A kind of improved method of the multitask concatenated convolutional neural network model for recognition of face |
| CN110263756A (en) * | 2019-06-28 | 2019-09-20 | 东北大学 | A kind of human face super-resolution reconstructing system based on joint multi-task learning |
| CN110647805A (en) * | 2019-08-09 | 2020-01-03 | 平安科技（深圳）有限公司 | Reticulated image identification method and device and terminal equipment |
| CN110647805B (en) * | 2019-08-09 | 2023-10-31 | 平安科技（深圳）有限公司 | Reticulate pattern image recognition method and device and terminal equipment |
| US20210141882A1 (en) * | 2019-11-12 | 2021-05-13 | Electronics And Telecommunications Research Institute | Apparatus and method for authenticating user based on multiple biometric information |
| US11829459B2 (en) * | 2019-11-12 | 2023-11-28 | Electronics And Telecommunications Research Institute | Apparatus and method for authenticating user based on multiple biometric information |
| CN111612133B (en) * | 2020-05-20 | 2021-10-19 | 广州华见智能科技有限公司 | Internal organ feature coding method based on face image multi-stage relation learning |
| CN111612133A (en) * | 2020-05-20 | 2020-09-01 | 广州华见智能科技有限公司 | Internal organ feature coding method based on face image multi-stage relation learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105760859B (en) | 2018-12-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105760859A (en) | Method and device for identifying reticulate pattern face image based on multi-task convolutional neural network | |
| CN109934200B (en) | RGB color remote sensing image cloud detection method and system based on improved M-Net | |
| CN106548159A (en) | Reticulate pattern facial image recognition method and device based on full convolutional neural networks | |
| CN105512638B (en) | A kind of Face datection and alignment schemes based on fusion feature | |
| CN103198493B (en) | A kind ofly to merge and the method for tracking target of on-line study based on multiple features self-adaptation | |
| CN104915676B (en) | SAR image sorting technique based on further feature study and watershed | |
| CN103870818B (en) | Smog detection method and device | |
| WO2020007307A1 (en) | Sky filter method for panoramic images and portable terminal | |
| CN106127204A (en) | A kind of multi-direction meter reading Region detection algorithms of full convolutional neural networks | |
| CN112288008B (en) | Mosaic multispectral image disguised target detection method based on deep learning | |
| CN104424457A (en) | Method for identifying two-dimensional code under the condition of nonlinear distortion | |
| CN106250895A (en) | A kind of remote sensing image region of interest area detecting method | |
| CN107909560A (en) | A kind of multi-focus image fusing method and system based on SiR | |
| CN103871039B (en) | Generation method for difference chart in SAR (Synthetic Aperture Radar) image change detection | |
| CN110188747A (en) | A kind of sloped correcting method of text image, device and image processing equipment | |
| Chen et al. | A blind CNN denoising model for random-valued impulse noise | |
| CN105590301B (en) | The Impulsive Noise Mitigation Method of adaptive just oblique diesis window mean filter | |
| CN110503613A (en) | Based on the empty convolutional neural networks of cascade towards removing rain based on single image method | |
| CN102682428A (en) | Fingerprint image computer automatic mending method based on direction fields | |
| CN108898562A (en) | A kind of mobile device image defogging method based on deep learning | |
| CN107945210A (en) | Target tracking algorism based on deep learning and environment self-adaption | |
| CN103927555B (en) | Static manual alphabet identifying system and method based on Kinect sensor | |
| CN113850725A (en) | Passive terahertz image target detection method for filtering enhanced deep learning | |
| CN105374047A (en) | Improved bilateral filtering and clustered SAR based image change detection method | |
| CN104778670A (en) | Fractal-wavelet self-adaption image denoising method based on multivariate statistical model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |