CN104268523A - Small-sample-based method for removing glasses frame in face image - Google Patents
Small-sample-based method for removing glasses frame in face image Download PDFInfo
- Publication number
- CN104268523A CN104268523A CN201410490947.5A CN201410490947A CN104268523A CN 104268523 A CN104268523 A CN 104268523A CN 201410490947 A CN201410490947 A CN 201410490947A CN 104268523 A CN104268523 A CN 104268523A
- Authority
- CN
- China
- Prior art keywords
- image
- matrix
- glasses frame
- frame
- facial image
- 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
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
- G06V40/162—Detection; Localisation; Normalisation using pixel segmentation or colour matching
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a small-sample-based method for removing a glasses frame in a face image. According to the method, no large numbers of samples or parameters are needed. The method comprises the steps that firstly, size and position normalization is carried out on the face image collected by an image collection device according to face key points, and a column vector is generated; then, a new matrix is formed by the column vector and a small sample face image training matrix prepared in advance, calculating and solving are carried out through a low-rank matrix, the column vector is decomposed into a column vector containing no glasses frame and an interference column vector containing the glasses frame; then, the interference column vector is reverted into the size of an original image; then, binarization processing is carried out on the image to obtain an image containing glasses frame position information; finally, the position of the glasses frame in the original image is filled with pixels through the difference value technology. In this way, the small-sample-based method for removing the glasses frame in the face image is implemented. According to the method, the correlation of face colors and the characteristics of the large glasses frame interference amplitude and the small number are sufficiently utilized, and the glasses frame in the face image can be automatically removed.
Description
Technical field
The present invention relates to a kind of method removing spectacle-frame in facial image based on small sample.
Background technology
Along with the popularization of face recognition technology, increasing occasion needs the application of face recognition technology, as occasions such as community gate inhibition, intelligent building, prison, bank vaults.But intelligent face recognition technology must remove some accessory information in facial image automatically, in order to avoid accessories bring impact to recognition of face, spectacle-frame is then one wherein.Secondly, traditional picture frame minimizing technology based on complexion model or geometric properties is not a priori in conjunction with the positional information of spectacle-frame, and have ignored some features that accessories have for face: the spectacle-frame i.e. pixel value amplitude difference that differs greatly compared with face is large, and spectacle-frame belongs to region among a small circle compared to whole face and has openness.
Summary of the invention
The object of the present invention is to provide a kind of method based on spectacle-frame in effective removal facial image of sample training, to improve the accuracy of recognition of face, calculated the positional information fully remaining facial image by low-rank matrix simultaneously, thus effectively extract the position of spectacle-frame, finally obtain the facial image after reducing.
In order to achieve the above object, method provided by the invention mainly comprises the following steps: 1) collecting 10-50 by arbitrary image collecting device opens the different facial images not wearing accessories in advance, then by all two dimensional images Zhang Chengyi dimensional vector respectively, and reformulate small sample facial image training matrix, the spectacle-frame for subsequent step is removed.2) spectacle-frame starting facial image is removed.The facial image collected by any collecting device positions according to the key point of facial image and carries out size and place normalization according to this locating information to face, and by two dimensional image Zhang Chengyi dimensional vector.3) column vector obtained and the small sample facial image training matrix obtained in advance are formed new matrix and solve interference matrix by low-rank matrix.4) from interference matrix, extract that column vector of corresponding facial image and be also reduced into original image size.5) image obtained is carried out binaryzation.The pixel of the picture frame position in the original image gathered user according to the image after binaryzation carries out difference and fills the image obtaining removing spectacle-frame.
Wherein, after low-rank matrix solves, the information comprised in interference matrix is utilized to go to judge the position of picture frame and non-reduced matrix.Because facial image training matrix has the characteristic of small sample, therefore, linear relationship between facial image is not strong, after solving, linear requirements can make the facial image after reducing have certain distortion and distortion by force, and the numerical value in interference matrix then well remains the composition of interference.The information that the present invention makes full use of part in interference is removed spectacle-frame, and unconventional method of reducing.
In sum, the present invention is solved by the low-rank matrix of small sample facial image training matrix, obtain the positional information of spectacle-frame, thus facial image is carried out to the method for picture frame removal, can effectively avoid because picture frame institute causes the shortcoming of trouble in human face recognition image, thus raising face identification rate.
Accompanying drawing explanation
Fig. 1 of the present inventionly removes the method for spectacle-frame in facial image based on small sample and is applied to the process flow diagram of the face identification system with image collecting device.
Fig. 2 is small sample training matrix and the column vector graph of a relation newly added.
Embodiment
Refer to that Fig. 1 is of the present invention to be removed the method for spectacle-frame in facial image based on small sample and be applied to the process flow diagram of the face identification system with image collecting device, it comprises the following steps.
The first step, collecting 10-50 by arbitrary image collecting device opens the different facial images not wearing accessories in advance, position (as: canthus, nose, the corners of the mouth, chin etc.) according to the key point of facial image, according to this locating information, size and place normalization are carried out to face, make the key point of each facial image substantially at same position.Then by all two dimensional images Zhang Chengyi dimensional vector respectively, and reformulate new matrix, wherein often arranging is all the dimensional vector that first toe-out becomes.This matrix is exactly small sample facial image training matrix, and the spectacle-frame for subsequent step is removed.
Second step, the spectacle-frame starting facial image is removed.The facial image collected by any collecting device positions according to the key point of facial image and carries out size and place normalization according to this locating information to face, makes the key point of each facial image substantially at same position.Then by two dimensional image Zhang Chengyi dimensional vector.
3rd step, the small sample facial image training matrix obtained in advance in the column vector obtained in second and the first step is formed new matrix, building form is consistent with the first step, new column vector splicing is after small sample facial image training matrix, and concrete pattern refers to Fig. 2 small sample training matrix and the column vector graph of a relation newly added.So far, obtain needs to process and the image array solved.
4th step, solve low-rank matrix according to formula (1), wherein D is the image array obtained in the 3rd step, and A is the facial image matrix (namely not containing the facial image matrix of the interference such as accessories) after reduction, E is isolated interference matrix, comprises the accessories interfere informations such as spectacle-frame in this matrix.Thus, obtain matrix A and E, wherein E is the matrix that user needs.Wherein why D can resolve into A and add E mainly because the row in this matrix meet following characteristics: 1) have certain linear relationship between each row, because facial image is have certain linear relationship and the broca scale picture of people has certain linear dependence in low-dimensional embedded space.Thus constraint rank (A) <r met in formula (1).2) interference is a small amount of and sparse for each row, this is because the interference of spectacle-frame is a small amount of for whole facial image, thus meets the constraint in formula (1) || E||0<epsilon.Therefore, user reasonably can solve A and E.The method solving low-rank matrix is existing a lot, and not in this method, so do not explain.
5th step, extracts that column vector of the facial image that corresponding user gathers and is also reduced into original image size from interference matrix E.
6th step, carry out binaryzation to the image obtained in the 5th step, wherein threshold value can be set as 20.According to the image after binaryzation, think those sizes be 1 pixel be the position at picture frame place, the pixel of the picture frame position in the original image therefore gathered user according to this information carries out difference and fills the image obtaining removing spectacle-frame.
In sum, of the present inventionly remove the personnel image of method to wearing spectacles frame of spectacle-frame in facial image based on small sample and carry out removing based on the spectacle-frame of sample, reached the effect of picture frame removal by a small amount of sample, thus visible ray, near infrared can be reached significant recognition effect and be promoted.
Formula 1:D=A+E s.t || E||0<epsilon, rank (A) <r.
Claims (2)
1. remove a method for spectacle-frame in facial image based on small sample, be applied to the face identification system with image collecting device, its key step comprises:
1) cut-and-dried small sample face training matrix, this matrix is opened the facial image not wearing accessories and is carried out critical area by by collecting 10-50 and locate the column vector that normalization posttension becomes and form;
2) facial image collected by any collecting device positions according to the key point of facial image and carries out size and place normalization according to this locating information to face, and by two dimensional image Zhang Chengyi dimensional vector;
Form low-rank matrix with cut-and-dried small sample face training matrix simultaneously;
3) solve interference matrix by low-rank matrix, from interference matrix, extract that column vector of corresponding facial image and be also reduced into original image size;
4) image obtained is carried out binaryzation;
The pixel of the picture frame position in the original image gathered user according to the image after binaryzation carries out difference and fills the image obtaining removing spectacle-frame.
2. a kind of method removing spectacle-frame in facial image based on small sample as claimed in claim 1, is characterized in that: the acquisition of spectacle-frame information is that the interference matrix solved by low-rank matrix is obtained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410490947.5A CN104268523A (en) | 2014-09-24 | 2014-09-24 | Small-sample-based method for removing glasses frame in face image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410490947.5A CN104268523A (en) | 2014-09-24 | 2014-09-24 | Small-sample-based method for removing glasses frame in face image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN104268523A true CN104268523A (en) | 2015-01-07 |
Family
ID=52160043
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410490947.5A Pending CN104268523A (en) | 2014-09-24 | 2014-09-24 | Small-sample-based method for removing glasses frame in face image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104268523A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107945126A (en) * | 2017-11-20 | 2018-04-20 | 杭州登虹科技有限公司 | Spectacle-frame removing method, device and medium in a kind of image |
| WO2018072102A1 (en) * | 2016-10-18 | 2018-04-26 | 华为技术有限公司 | Method and apparatus for removing spectacles in human face image |
| CN108076290A (en) * | 2017-12-20 | 2018-05-25 | 维沃移动通信有限公司 | A kind of image processing method and mobile terminal |
| CN108765289A (en) * | 2018-05-25 | 2018-11-06 | 李锐 | A kind of extraction splicing of digital picture and reduction fill method |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090180671A1 (en) * | 2007-10-19 | 2009-07-16 | Samsung Electronics Co., Ltd. | Multi-view face recognition method and system |
| CN103632134A (en) * | 2013-10-17 | 2014-03-12 | 浙江师范大学 | Human face identification method based on fisher low-rank matrix restoration |
-
2014
- 2014-09-24 CN CN201410490947.5A patent/CN104268523A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090180671A1 (en) * | 2007-10-19 | 2009-07-16 | Samsung Electronics Co., Ltd. | Multi-view face recognition method and system |
| CN103632134A (en) * | 2013-10-17 | 2014-03-12 | 浙江师范大学 | Human face identification method based on fisher low-rank matrix restoration |
Non-Patent Citations (1)
| Title |
|---|
| YANGMUZI ZHANG，AND ETC: "Learning Structured Low-rank Representations for Image Classification", 《2013 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018072102A1 (en) * | 2016-10-18 | 2018-04-26 | 华为技术有限公司 | Method and apparatus for removing spectacles in human face image |
| CN107945126A (en) * | 2017-11-20 | 2018-04-20 | 杭州登虹科技有限公司 | Spectacle-frame removing method, device and medium in a kind of image |
| CN108076290A (en) * | 2017-12-20 | 2018-05-25 | 维沃移动通信有限公司 | A kind of image processing method and mobile terminal |
| CN108765289A (en) * | 2018-05-25 | 2018-11-06 | 李锐 | A kind of extraction splicing of digital picture and reduction fill method |
| CN108765289B (en) * | 2018-05-25 | 2022-02-18 | 李锐 | Digital image extracting, splicing, restoring and filling method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107220611B (en) | Space-time feature extraction method based on deep neural network | |
| CN103020965B (en) | A kind of foreground segmentation method based on significance detection | |
| CN107133969B (en) | A kind of mobile platform moving target detecting method based on background back projection | |
| CN104268523A (en) | Small-sample-based method for removing glasses frame in face image | |
| CN103679147A (en) | Method and device for identifying model of mobile phone | |
| CN109509164A (en) | A kind of Multisensor Image Fusion Scheme and system based on GDGF | |
| CN103400155A (en) | Pornographic video detection method based on semi-supervised learning of images | |
| CN103327359B (en) | Video significance region searching method applied to video quality evaluation | |
| CN105894513B (en) | Take the remote sensing image variation detection method and system of imaged object change in time and space into account | |
| CN103530638A (en) | Method for matching pedestrians under multiple cameras | |
| CN109785321A (en) | Meibomian gland method for extracting region based on deep learning and Gabor filter | |
| CN106228157A (en) | Coloured image word paragraph segmentation based on image recognition technology and recognition methods | |
| CN110458115A (en) | A kind of integrated algorithm of target detection of the multiframe based on timing | |
| CN109886170A (en) | A kind of identification of oncomelania intelligent measurement and statistical system | |
| CN103108124A (en) | Image acquisition method, device and mobile terminal | |
| CN106485698A (en) | The method obtaining DNA chromatograph from gel electrophoresis strip image | |
| CN109961016A (en) | The accurate dividing method of more gestures towards Intelligent household scene | |
| CN110826612A (en) | Training and identifying method for deep learning | |
| CN104239883A (en) | Textural feature extraction method and device | |
| CN103824067B (en) | The location of a kind of image main target and recognition methods | |
| CN101763513A (en) | Foreground extraction method for removing light effect | |
| CN103605959A (en) | A method for removing light spots of iris images and an apparatus | |
| CN102799855B (en) | Based on the hand positioning method of video flowing | |
| CN104063879B (en) | Pedestrian flow estimation method based on flux and shielding coefficient | |
| CN106921893A (en) | A kind of advertisement sending method based on age bracket |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20150107 |