CN102110303A - Method for compounding face fake portrait\fake photo based on support vector return - Google Patents
Method for compounding face fake portrait\fake photo based on support vector return Download PDFInfo
- Publication number
- CN102110303A CN102110303A CN 201110058175 CN201110058175A CN102110303A CN 102110303 A CN102110303 A CN 102110303A CN 201110058175 CN201110058175 CN 201110058175 CN 201110058175 A CN201110058175 A CN 201110058175A CN 102110303 A CN102110303 A CN 102110303A
- Authority
- CN
- China
- Prior art keywords
- portrait
- photo
- pseudo
- piece
- support vector
- 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
Images
Landscapes
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method for compounding face fake portrait\fake photo based on support vector return. The method comprises steps as follows: (1) partitioning database sample set; (2) randomly taking a photo\portrait from a photo\portrait testing set; (3) generating fake portrait\fake photo initial estimate corresponding to the randomly-taken photo\portrait; (4) generating a training face photo\portrait block set; (5) generating a support vector return model between the training face photo\portrait block set and a training face portrait/photo block set; (6) generating fake portrait\fake photo high frequency corresponding to the randomly-taken photo\portrait; and (7) generating the final fake portrait\fake photo. The invention combines the initial estimate and the high frequency to generate the fake portrait\fake photo, so as to enable the generated fake portrait\fake photo to be clearer and improve the discrimination when searching the fake portrait\fake photo. The invention adopts the method based on support vector return, and the method can also be used in small sample issues.
Description
Technical field
The invention belongs to technical field of image processing, further relate to a kind of synthetic method that adopts the human face portrait-photo of pattern-recognition and computer vision technique, can be used for the identification of human face portrait-photo and portrait in the entertainment field or photo generation in the criminal investigation field.
Background technology
At present in criminal investigation and entertainment field; any human face photo is generated a pseudo-portrait or any human face portrait is generated a pseudo-photo usually based on three kinds of methods: based on local linear human face portrait synthetic method, based on the human face portrait synthetic method of built-in type hidden Markov model with based on the human face portrait-picture synthesis method of rarefaction representation.
People such as Liu have proposed a kind of local linearity and have been similar to non-linear method and generate pseudo-portrait in document " Q.S.Liu and X.O.Tang; A nonlinear approach for face sketch synthesis and recognition; in Proc.IEEE Int.Conference on Computer Vision; San Diego; CA; pp.1005-1010,20-26Jun.2005. ".This method is for any photo, at first it is carried out piecemeal, for any one fritter behind the piecemeal, obtain the fitting coefficient of its most similar K photo piece and this K photo piece with the linear embedding grammar in part, in tranining database, seek and this K the portrait piece that the photo piece is corresponding then, the coefficient that the match of photo piece is obtained makes up with the portrait piece, obtains final puppet portrait at last.Because this method adopted piecemeal k nearest neighbor technology, and the last overlapping part of piecemeal will average, thereby causes detailed information to be lost and then reduced human face portrait-photo discrimination.
People such as Gao propose to utilize the built-in type hidden Markov model to generate pseudo-portrait in document " Gao, X., Zhong; J.; Tao, T.and Li, X.; Local face sketch synthesis learning; Neurocomputing, vol.71, no.10-12; pp.1921-1930, Jun.2008. ".This method is at first carried out piecemeal to photo and portrait in the training storehouse, with the built-in type hidden Markov model corresponding photo piece and portrait piece are carried out modeling then, give a photo arbitrarily, carry out piecemeal equally, for piece arbitrarily, with the integrated thought of selectivity, thereby the model that the selection portion piecemeal generates carries out the generation of pseudo-portrait and merges obtaining final puppet portrait.Because this method adopted the selectivity integrated technology, the puppet of generation portrait will be weighted average, causes detailed information to be lost and then has reduced portrait-photo discrimination, and the algorithm complex of this method is too high in addition.
People such as high-new ripple are at application number: 201010289330.9 applyings date: disclosed in the patent application document of 2010-09-24 publication number: 101958000A " based on the portrait-photograph generation method of rarefaction representation " strengthens the detailed information of puppet portrait and pseudo-photo.This method at first generates the initial estimation of pseudo-portrait or pseudo-photo with the method for local linear synthetic method or built-in type hidden Markov model, use then based on the synthetic pseudo-portrait of method of rarefaction representation or the detailed information of pseudo-photo, at last initial estimation and detailed information are merged.But the method for this patented claim is inapplicable for the less small sample problem of sample number.
In sum, existing portrait-photo composition algorithm all exists detailed information to lose and the fuzzy problem that causes.Lose and cause portrait-photo discrimination lower owing to detailed information during simultaneously with the synthetic pseudo-photo of existing method and pseudo-portrait the-photo retrieval.The used sample number of the method for existing in addition enhancing details is bigger, makes inapplicable for the less small sample problem of sample number.
Summary of the invention
The objective of the invention is to overcome the deficiency of above-mentioned prior art, propose a kind of synthetic method of the human face portrait-photo that returns based on support vector, can be criminal investigation field or entertainment field pseudo-clearly portrait and pseudo-photo are provided.
The concrete steps that the present invention realizes are as follows:
(1) dividing data storehouse sample set: with pending human face photo portrait collection be divided into respectively photo portrait training set and photo the portrait test set;
(2) photo appoint in the portrait test set get a photo portrait;
(3) generate pseudo-portrait pseudo-photo initial estimation: to the photo got portrait, with embedded method generate corresponding puppet portrait puppet photo initial estimation;
(4) generate one group of training Ren Lianzhaopian portrait piece collection:
4a) photo and portrait branch squarely photo piece and the portrait piece that will train photograph collection and training portrait to concentrate generates Zhao Pian portrait piece collection;
4b) comparison film each piece of concentrating of portrait piece, ask the first order derivative and the second derivative of horizontal direction and vertical direction respectively, and with derivative value as column vector;
4c) column vector that generates with C means clustering algorithm Dui Zhaopian portrait piece is carried out cluster, obtains one group of training Ren Lianzhaopian portrait piece collection, for each group Zhao Pian portrait piece collection is set corresponding sequence number class mark;
4d) will divide same concentrated, generate one group of training Ren Lianhuaxiang photo piece collection corresponding to Zhao Pian portrait piece collection same item target portrait photo piece;
(5) generate one group of support vector regression model:
5a) for step 4c) one group of training Ren Lianzhaopian portrait piece collection, be in line after all pixel values of each piece are deducted the piece center pixel value, all row of all pieces form a line as the input of support vector regression model;
5b) for step 4c) one group of training Ren Lianhuaxiang photo piece collection, the center pixel value of each piece is deducted the value that the mean value of all pixels in its piece obtains forms a line as the output of support vector regression model;
5c) at step 5a) in photo portrait piece the collection input and the corresponding step 5b that generate) in portrait one group of support vector regression model of training generation between the output that generates of photo piece collection;
(6) generate to appoint get photo the corresponding pseudo-portrait of portrait the high frequency of pseudo-photo:
The Zhao Pian portrait branch squarely fritter of 6a) step (2) Zhao Pian portrait test set being got;
6b) to step 4c) in each training photo of generating portrait piece collection calculating mean value;
6c) to step 6a) in each fritter, calculate itself and step 6b) in each photo Euclidean distance between the portrait piece ensemble average value, obtain nearest photo portrait piece collection, according to step 5c) take out this photo the portrait piece set pair support vector regression model of answering;
6d) with step 6a) in be in line as importing after all pixel values deduct the piece center pixel value in the piece of each fritter, by step 6c) in the support vector regression model obtain output valve, with its output valve as the puppet of fritter portrait pseudo-photo high frequency, with the puppet portrait of all fritters pseudo-photo high frequency combination obtain final pseudo-portrait pseudo-photo high frequency;
(7) generate final puppet portrait pseudo-photo: with the puppet portrait that generates in the step (3) the puppet portrait that generates in pseudo-photo initial estimation and the step (6) the addition of pseudo-photo high frequency obtain final puppet portrait pseudo-photo.
Compared with prior art, the present invention has following advantage:
(1) the present invention adopted puppet portrait that the method for initial estimation and high-frequency information addition make to generate pseudo-photo more clear;
(2) the present invention when synthetic high-frequency information, adopt the method for the synthetic high-frequency information of piecemeal cluster make the final puppet portrait that generates pseudo-photo discrimination when retrieving higher;
(3) the present invention adopts the synthetic high-frequency information of method of support vector recurrence to make when solving small sample problem suitable equally.
Technical process of the present invention and effect can describe in detail in conjunction with the following drawings.
Description of drawings
Fig. 1 is a process flow diagram of the present invention;
The design sketch of the puppet portrait that Fig. 2 generates on CUHK student database for the present invention;
The design sketch of the pseudo-photo that Fig. 3 generates on CUHK student database for the present invention.
Embodiment
With reference to Fig. 1, the specific embodiment of the invention is as follows:
Step 1, dividing data storehouse sample set.
With the human face photo in the pending CUHK student database portrait collection be divided into respectively photo portrait training set U={U
p| U
p∈ R
nAnd photo the portrait test set
Make in the database sample set training set and
The ratio of test set is 40%～60%.
Wherein, U
pThe expression photo the p of portrait in the training set open photo portrait,
P=1,2 ..., M, M=100, M are the numbers of training sample set,
R
nExpression n dimension real number space,
U
qThe expression photo the q of portrait in the test set open photo portrait,
Q=1 ..., N, N=88, N are the numbers of test sample book collection,
Step 2, appoint get a photo portrait.For generate pseudo-portrait pseudo-photo initial estimation and pseudo-portrait pseudo-photo high frequency, the photo that need obtain in step (1) the portrait test set
In appoint get a photo portrait.
Step 3, initial estimation.
Comparison film the portrait test set
Appoint the photo get portrait, with embedded method generate its correspondence the puppet portrait puppet photo initial estimation L
q, embedded method can adopt local linear (LLE) method or built-in type hidden Markov model (E-HMM) method of embedding.
Step 4 generates photo piece collection and portrait piece collection.
4a) piecemeal.
Will training photo that step 1 obtains among the portrait collection U photo portrait by square photo piece and the portrait piece of being divided into.The lap size is 3/4 of a square block size.The method of partition that the embodiment of the invention adopts is, 64 * 64 photo and portrait that training photograph collection and training portrait are concentrated, at first the edge is all expanded 3 pixels, 64 * 64 photo and portrait become 70 * 70, block size is taken as 7 * 7, lap is 5 * 5, thereby every photo can be divided into 1024 squares with portrait, and the square number of human face photo piece collection and human face portrait piece collection all is 102400;
4b) to the training photo each piece among the portrait collection U, respectively get first order derivative and second derivative in the horizontal direction with on the vertical direction, the linearity of first order derivative and second derivative is extracted operator and is respectively f
1=[1,0,1] and f
2=[1,0 ,-2,0,1] then, forms a line four derivative value as column vector again;
4c) column vector is carried out cluster with the C means clustering algorithm, cluster class number generally gets 10～100 by experiment, cluster class number adopts 25 in the embodiments of the invention, promptly obtains 25 groups of training Ren Lianzhaopian portrait piece collection, again each group Zhao Pian portrait piece collection is set corresponding sequence number class mark;
4d) according to photo portrait piece collection sequence number class mark, for step 4a) in the portrait that generates the photo piece, seek the class mark of the affiliated training Ren Lianzhaopian portrait piece collection of its corresponding Zhao Pian portrait piece, to belong to same class target portrait photo piece and be divided into one group, generate 25 groups of training Ren Lianhuaxiang photo piece collection.
Step 5 generates one group of support vector regression model.
5a) for step 4c) 25 groups of training Ren Lianzhaopian portrait piece collection obtaining, be in line after all pixel values of each piece are deducted the piece center pixel value, all row of all pieces form a line as the input of support vector regression model;
5b) for step 4c) 25 groups of training Ren Lianhuaxiang photo piece collection obtaining, the center pixel value of each piece is deducted the value that the mean value of all pixels in its piece obtains form a line as the output of support vector regression model;
5c) at step 5a) in photo portrait piece the collection input and the corresponding step 5b that generate) in portrait 25 groups of support vector regression models of training generation between the output that generates of photo piece collection.The LIBSVM kit of Taiwan Univ.'s woods intelligence benevolence is adopted in the realization of support vector regression model.
Step 6, generate pseudo-portrait pseudo-photo high frequency.
6a) with step 2 photo a photo being got in the test set of portrait portrait U
q *Divide the squarely fritter, for 64 * 64 Zhao Pian portrait, lap lacks a pixel than square block.The method of partition that the embodiment of the invention adopts is, at first the edge all expanded 3 pixels, such 64 * 64 photo portrait just become 70 * 70 photo portrait, get block size then and be taken as 7 * 7, lap is 6 * 6, the test photo portrait U
q *The piece number be 4096;
6b) to step 4c) in each training photo of generating portrait piece collection calculating mean value;
6c) to step 6a) in each fritter, calculate itself and step 6b) in each photo Euclidean distance between the portrait piece ensemble average value, obtain nearest photo portrait piece collection, take out this photo portrait piece collection at step 5c) in the support vector regression model of correspondence;
6d) with step 6a) in be in line as importing after all pixel values deduct the piece center pixel value in the piece of each fritter, by step 6c) in the support vector regression model obtain output valve, with its output valve as the puppet of fritter portrait pseudo-photo high frequency, with the puppet portrait of all fritters pseudo-photo high frequency combination obtain final pseudo-portrait pseudo-photo high frequency.
Step 7, generate final puppet portrait pseudo-photo.
With the puppet portrait that generates in the step 3 pseudo-photo initial estimation L
qWith the puppet portrait that generates in the step 6 pseudo-photo high frequency H
qAddition obtain final puppet portrait pseudo-photo.
Effect of the present invention can further specify by following emulation experiment.
1. experiment condition
The computer configuration environment is Pentinum (R) 1.73Ghz, internal memory 1G, and system windows XP, simulation software adopts MATLAB R2007a.Database adopts the CUHK student of Hong Kong Chinese University database.SVR realizes adopting the MTTLAB kit " http://www.csie.ntu.edu.tw/～cjlin/libsvm/ " of Taiwan Univ.'s woods intelligence benevolence.
2. experiment content
The present invention has 3 groups of experiments:
Experiment 1: synthetic pseudo-portrait on CUHK student database.
Experiment 2: synthetic pseudo-photo on CUHK student database.
Experiment 3: the inventive method is on puppet portrait and pseudo-photo that experiment 1 and experiment 2 are synthesized, and utilization is carried out the experiment that human face photo-portrait is discerned based on the face identification method of rarefaction representation.
3. experimental result and analysis
3.1 pseudo-portrait is synthetic
LLE is returned the synthetic method note of (SVR) synthetic high frequency with support vector then as initial estimation make SVR-LLE, E-HMM is made SVR-E-HMM with the method note of the synthetic high frequency of support vector recurrence (SVR) then as initial estimation.For support vector homing method of the present invention, be input as the photo that test data is concentrated, Fig. 2 has shown the output result of pseudo-portrait on the CUHKstudent database.In Fig. 2, Fig. 2 (a) is the original photo of CUHK student database, Fig. 2 (b) is original portrait, Fig. 2 (c) is the puppet portrait that generates with LLE, Fig. 2 (d) is the puppet portrait that generates with E-HMM, Fig. 2 (e) is the puppet portrait that generates with SVR-LLE, and Fig. 2 (f) is the puppet portrait that generates with SVR-E-HMM; As can be seen from Figure 2, (e) that is generated by the present invention and (c) that (f) generate than existing method respectively and (d) more clear can be used for amusement and draw a portrait with the photo generation.
3.2 pseudo-photo is synthetic
LLE is returned the synthetic method note of (SVR) synthetic high frequency with support vector then as initial estimation make SVR-LLE, E-HMM is made SVR-E-HMM with the method note of the synthetic high frequency of support vector recurrence (SVR) then as initial estimation.For support vector homing method of the present invention, be input as the portrait that test data is concentrated, Fig. 4 has shown the output result of pseudo-portrait on the CUHKstudent database.In Fig. 3, Fig. 3 (a) is the original photo of CUHK student database, Fig. 3 (b) is original portrait, Fig. 3 (c) is the pseudo-photo that generates with LLE, Fig. 3 (d) is the pseudo-photo that generates with E-HMM, Fig. 3 (e) is the pseudo-photo that generates with SVR-LLE, and Fig. 3 (f) is the pseudo-photo that generates with SVR-E-HMM; As can be seen from Figure 3, by (e) of the present invention's generation with (f) respectively than original method (c) and (d) more clear, can be used for amusement and generate photo with portrait.
3.3 human face photo-portrait identification based on rarefaction representation
Recognition of face mainly is used in public security organ and carries out suspect's retrieval.Appoint the portrait of giving a suspect, this portrait is retrieved from the picture data storehouse of public security organ.A kind of method is based on the method for pseudo-portrait, method from photo to portrait generates the puppet portrait of all photo correspondences the picture data storehouse, so just can retrieve from puppet portrait database this suspect's portrait, this system is called the people's face searching system based on the puppet portrait; Another kind method is based on the method for pseudo-photo, generate suspect and draw a portrait corresponding pseudo-photo from drawing a portrait the method for photo, so just can retrieve from the picture data storehouse suspect's pseudo-photo, this system is called the people's face searching system based on pseudo-photo.For the experiment of carrying out on the CUHKstudent database, when generating pseudo-portrait and pseudo-photo, 88 people of 188 philtrums are as training, and 100 people test generates corresponding puppet portrait or pseudo-photo.When using people's face searching system of drawing a portrait based on puppet to discern, 100 pseudo-portraits are done training, and 100 portraits are done test; When using based on the identification of people's face searching system type of pseudo-photo, 100 photos are done training, and 100 pseudo-photos are done test.Table 1 has shown based on people's face searching system of puppet portrait with based on the result for retrieval of people's face searching system on CUHK student database of pseudo-photo.As can be seen from Table 1, the portrait-picture synthesis method SVR-LLE that returns based on support vector of the present invention is than the discrimination height of classic method LLE.
Table 1. on CUHK student database, use based on puppet portrait the discrimination of people's face searching system of pseudo-photo
Claims (8)
1. the pseudo-picture synthesis method of the pseudo-portrait of people's face that returns based on support vector comprises the steps:
(1) dividing data storehouse sample set: with pending human face photo portrait collection be divided into respectively photo portrait training set and photo the portrait test set;
(2) photo appoint in the portrait test set get a photo portrait;
(3) generate pseudo-portrait pseudo-photo initial estimation: to the photo got portrait, with embedded method generate corresponding puppet portrait puppet photo initial estimation;
(4) generate one group of training Ren Lianzhaopian portrait piece collection:
4a) photo and portrait branch squarely photo piece and the portrait piece that will train photograph collection and training portrait to concentrate generates Zhao Pian portrait piece collection;
4b) comparison film each piece of concentrating of portrait piece, ask the first order derivative and the second derivative of horizontal direction and vertical direction respectively, and with derivative value as column vector;
4c) column vector that generates with C means clustering algorithm Dui Zhaopian portrait piece is carried out cluster, obtains one group of training Ren Lianzhaopian portrait piece collection, for each group Zhao Pian portrait piece collection is set corresponding sequence number class mark;
4d) will divide same concentrated, generate one group of training Ren Lianhuaxiang photo piece collection corresponding to Zhao Pian portrait piece collection same item target portrait photo piece;
(5) generate one group of support vector regression model:
5a) for step 4c) one group of training Ren Lianzhaopian portrait piece collection, be in line after all pixel values of each piece are deducted the piece center pixel value, all row of all pieces form a line as the input of support vector regression model;
5b) for step 4c) one group of training Ren Lianhuaxiang photo piece collection, the center pixel value of each piece is deducted the value that the mean value of all pixels in its piece obtains forms a line as the output of support vector regression model;
5c) at step 5a) in photo portrait piece the collection input and the corresponding step 5b that generate) in portrait one group of support vector regression model of training generation between the output that generates of photo piece collection;
(6) generate to appoint get photo the corresponding pseudo-portrait of portrait the high frequency of pseudo-photo:
The Zhao Pian portrait branch squarely fritter of 6a) step (2) Zhao Pian portrait test set being got;
6b) to step 4c) in each training photo of generating portrait piece collection calculating mean value;
6c) to step 6a) in each fritter, calculate itself and step 6b) in each photo Euclidean distance between the portrait piece ensemble average value, obtain nearest photo portrait piece collection, according to step 5c) take out this photo the portrait piece set pair support vector regression model of answering;
6d) with step 6a) in be in line as importing after all pixel values deduct the piece center pixel value in the piece of each fritter, by step 6c) in the support vector regression model obtain output valve, with its output valve as the puppet of fritter portrait pseudo-photo high frequency, with the puppet portrait of all fritters pseudo-photo high frequency combination obtain final pseudo-portrait pseudo-photo high frequency;
(7) generate final puppet portrait pseudo-photo: with the puppet portrait that generates in the step (3) the puppet portrait that generates in pseudo-photo initial estimation and the step (6) the addition of pseudo-photo high frequency obtain final puppet portrait pseudo-photo.
2. the pseudo-portrait of people's face that returns based on support vector according to claim 1 pseudo-picture synthesis method, it is characterized in that: described step (1) divide human face photo during the portrait collection, the proportional range of training set and test set is 2/5～3/5 in the database sample set.
3. the pseudo-picture synthesis method of the pseudo-portrait of people's face that returns based on support vector according to claim 1, it is characterized in that: the embedded method that described step (3) adopts is local linear embedding grammar.
4. the pseudo-portrait of people's face that returns based on support vector according to claim 1 pseudo-picture synthesis method, it is characterized in that: the embedded method that described step (3) adopts is a built-in type hidden Markov model method.
5. the pseudo-picture synthesis method of the pseudo-portrait of people's face that returns based on support vector according to claim 1, it is characterized in that: in the branch squarely fritter described step 4a), the lap size is 3/4 of a square block size.
6. the pseudo-portrait of people's face that returns based on support vector according to claim 1 pseudo-picture synthesis method, it is characterized in that: the first order derivative described step 4b) and the linearity of second derivative are extracted operator and are respectively f
1=[1,0,1] and f
2=[1,0 ,-2,0,1].
7. the pseudo-portrait of people's face that returns based on support vector according to claim 1 pseudo-picture synthesis method, it is characterized in that: the span of the cluster class number described step 4c) is 10～100.
8. the pseudo-picture synthesis method of the pseudo-portrait of people's face that returns based on support vector according to claim 1, it is characterized in that: the branch squarely fritter described step 6a), lap lacks a pixel than square block.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2011100581754A CN102110303B (en) | 2011-03-10 | 2011-03-10 | Method for compounding face fake portrait\fake photo based on support vector return |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2011100581754A CN102110303B (en) | 2011-03-10 | 2011-03-10 | Method for compounding face fake portrait\fake photo based on support vector return |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102110303A true CN102110303A (en) | 2011-06-29 |
| CN102110303B CN102110303B (en) | 2012-07-04 |
Family
ID=44174450
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2011100581754A Expired - Fee Related CN102110303B (en) | 2011-03-10 | 2011-03-10 | Method for compounding face fake portrait\fake photo based on support vector return |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102110303B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103984954A (en) * | 2014-04-23 | 2014-08-13 | 西安电子科技大学宁波信息技术研究院 | Image synthesis method based on multi-feature fusion |
| CN104517274A (en) * | 2014-12-25 | 2015-04-15 | 西安电子科技大学 | Face portrait synthesis method based on greedy search |
| CN104700380A (en) * | 2015-03-12 | 2015-06-10 | 陕西炬云信息科技有限公司 | Face portrait compositing method based on single photos and portrait pairs |
| CN108124488A (en) * | 2017-12-12 | 2018-06-05 | 福建联迪商用设备有限公司 | A kind of payment authentication method and terminal based on face and vocal print |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101169830A (en) * | 2007-11-30 | 2008-04-30 | 西安电子科技大学 | Human face portrait automatic generation method based on embedded type hidden markov model and selective integration |
| CN101216889A (en) * | 2008-01-14 | 2008-07-09 | 浙江大学 | A face image super-resolution method with the amalgamation of global characteristics and local details information |
| KR100974165B1 (en) * | 2008-02-15 | 2010-08-04 | 한양대학교 산학협력단 | System and method for synthesis of face expression using nonlinear regression analysis |
-
2011
- 2011-03-10 CN CN2011100581754A patent/CN102110303B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101169830A (en) * | 2007-11-30 | 2008-04-30 | 西安电子科技大学 | Human face portrait automatic generation method based on embedded type hidden markov model and selective integration |
| CN101216889A (en) * | 2008-01-14 | 2008-07-09 | 浙江大学 | A face image super-resolution method with the amalgamation of global characteristics and local details information |
| KR100974165B1 (en) * | 2008-02-15 | 2010-08-04 | 한양대학교 산학협력단 | System and method for synthesis of face expression using nonlinear regression analysis |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103984954A (en) * | 2014-04-23 | 2014-08-13 | 西安电子科技大学宁波信息技术研究院 | Image synthesis method based on multi-feature fusion |
| CN104517274A (en) * | 2014-12-25 | 2015-04-15 | 西安电子科技大学 | Face portrait synthesis method based on greedy search |
| CN104517274B (en) * | 2014-12-25 | 2017-06-16 | 西安电子科技大学 | Human face portrait synthetic method based on greedy search |
| CN104700380A (en) * | 2015-03-12 | 2015-06-10 | 陕西炬云信息科技有限公司 | Face portrait compositing method based on single photos and portrait pairs |
| CN104700380B (en) * | 2015-03-12 | 2017-08-15 | 陕西炬云信息科技有限公司 | Based on single photo with portrait to human face portrait synthetic method |
| CN108124488A (en) * | 2017-12-12 | 2018-06-05 | 福建联迪商用设备有限公司 | A kind of payment authentication method and terminal based on face and vocal print |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102110303B (en) | 2012-07-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Tu et al. | Edge-guided non-local fully convolutional network for salient object detection | |
| Cheng et al. | Perturbation-seeking generative adversarial networks: A defense framework for remote sensing image scene classification | |
| CN109359538B (en) | Training method of convolutional neural network, gesture recognition method, device and equipment | |
| Kliper-Gross et al. | Motion interchange patterns for action recognition in unconstrained videos | |
| CN110647817A (en) | Real-time face detection method based on MobileNet V3 | |
| CN110543841A (en) | Pedestrian re-identification method, system, electronic device and medium | |
| CN109064502A (en) | The multi-source image method for registering combined based on deep learning and artificial design features | |
| CN103279936A (en) | Human face fake photo automatic combining and modifying method based on portrayal | |
| Qiang et al. | SqueezeNet and fusion network-based accurate fast fully convolutional network for hand detection and gesture recognition | |
| Zhang et al. | Random reconstructed unpaired image-to-image translation | |
| WO2004008392A1 (en) | Image matching system using 3-dimensional object model, image matching method, and image matching program | |
| CN104463250A (en) | Sign language recognition translation method based on Davinci technology | |
| CN101958000B (en) | Face image-picture generating method based on sparse representation | |
| CN105718873A (en) | People stream analysis method based on binocular vision | |
| CN106529547B (en) | A kind of Texture Recognition based on complete local feature | |
| WO2021169257A1 (en) | Face recognition | |
| CN104077742B (en) | Human face sketch synthetic method and system based on Gabor characteristic | |
| Yi et al. | Motion keypoint trajectory and covariance descriptor for human action recognition | |
| CN102110303B (en) | Method for compounding face fake portrait\fake photo based on support vector return | |
| Kihl et al. | A unified framework for local visual descriptors evaluation | |
| CN103714331A (en) | Facial expression feature extraction method based on point distribution model | |
| Neverova | Deep learning for human motion analysis | |
| Choi et al. | Data insufficiency in sketch versus photo face recognition | |
| CN114463805B (en) | Deep forgery detection method, device, storage medium and computer equipment | |
| JP4570995B2 (en) | MATCHING METHOD, MATCHING DEVICE, AND PROGRAM |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20120704 Termination date: 20180310 |