CN106682587A - Image database building method and device - Google Patents
Image database building method and device Download PDFInfo
- Publication number
- CN106682587A CN106682587A CN201611099286.9A CN201611099286A CN106682587A CN 106682587 A CN106682587 A CN 106682587A CN 201611099286 A CN201611099286 A CN 201611099286A CN 106682587 A CN106682587 A CN 106682587A
- Authority
- CN
- China
- Prior art keywords
- image
- width
- width image
- predetermined patterns
- alignment
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/51—Indexing; Data structures therefor; Storage structures
-
- 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
- G06V40/179—Human faces, e.g. facial parts, sketches or expressions metadata assisted face recognition
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Processing Or Creating Images (AREA)
- Image Processing (AREA)
Abstract
The invention can be applied to the technical field of digital image processing, and provides an image database building method and device. The image database building method comprises the steps of carrying out alignment operation on N images, wherein N is an integer larger than 1, conducting key point detection of preset positions in each of the aligned N images, positioning and extracting regional images in the preset positions in the N images according to detection results of key points, interchanging the regional images in the same preset positions in the N images and then saving new images into the image database. By aligning the N images, detecting the key points in the preset positions, extracting the regional images of the preset positions, interchanging the regional images and then synthesizing the new images, the image database building method and device can generate a large-sample-size image database based on few image samples.
Description
Technical field
The invention belongs to digital image processing techniques field, more particularly to a kind of image base construction method and device.
Background technology
When recognizing face and its attribute using the method for deep learning, the quantity of image data base directly affects identification standard
True rate.For oversampler method, oversampler method is according to upper and lower to piece image to the image base construction method for mainly adopting at present
Totally 5 directions are intercepted for left and right and center, preserve the image and its mirror image figure, and 1 image can expand to 10 images.But
Be the image library that this method builds amount of images it is few so that during method identification face and its attribute based on deep learning
Accuracy rate it is not high.
The content of the invention
The purpose of the embodiment of the present invention is to provide a kind of image base construction method and device, it is intended to solve what is built at present
The few problem of image library amount of images.
The embodiment of the present invention is achieved in that a kind of image base construction method, including：
Alignment operation is carried out to N width image, the N is the integer more than 1；
The N width image after to alignment carries out respectively the critical point detection of predetermined patterns；
According to the result of the critical point detection, the region of the predetermined patterns is positioned and extracted in the N width image
Image；
The area image that the identical predetermined patterns are carried out in the N width image is exchanged, and the new images for obtaining are deposited
Enter image library.
The another object of the embodiment of the present invention is to provide a kind of image library construction device, including：
Alignment unit, for carrying out alignment operation to N width image, the N is the integer more than 1；
Detector unit, to alignment after the N width image carry out the critical point detection of predetermined patterns respectively；
Extraction unit, for according to the result of the critical point detection, default portion being positioned and extracting in the N width image
The area image of position；
Exchange unit, the area image for carrying out the identical predetermined patterns in the N width image is exchanged, and will
To new images be stored in image library.
In embodiments of the present invention, operated by image alignment, predetermined patterns critical point detection, extract the area of predetermined patterns
Area image, carries out area image exchange, so as to synthesize new image, based on a small amount of image pattern large sample amount just can be generated
Image library.
Description of the drawings
Fig. 1 is the flow chart of image base construction method provided in an embodiment of the present invention；
Fig. 2 is that image base construction method S101 provided in an embodiment of the present invention implements flow chart；
Fig. 3 is that image base construction method S104 provided in an embodiment of the present invention implements flow chart；
Fig. 4 A are the schematic diagrams before and after exchange face provided in an embodiment of the present invention；
Fig. 4 B are the schematic diagrams before and after exchange left eye provided in an embodiment of the present invention；
Fig. 4 C are the schematic diagrams before and after exchange right eye provided in an embodiment of the present invention；
Fig. 4 D are the schematic diagrams before and after exchange nose provided in an embodiment of the present invention；
Fig. 4 E are the schematic diagrams exchanged before and after remaining face provided in an embodiment of the present invention；
Fig. 5 is the schematic diagram that 5 people provided in an embodiment of the present invention synthesize the partial synthesis design sketch in 100 images；
Fig. 6 is the parts of images schematic diagram in third-party testing storehouse provided in an embodiment of the present invention；
Fig. 7 is the structured flowchart of image library construction device provided in an embodiment of the present invention.
Specific embodiment
In order that the objects, technical solutions and advantages of the present invention become more apparent, it is right below in conjunction with drawings and Examples
The present invention is further elaborated.It should be appreciated that specific embodiment described herein is used only for explaining the present invention, and
It is not used in the restriction present invention.
The embodiment of the present invention is introduced by taking facial image as an example, but the image library for building includes but is not limited to facial image
Storehouse, can also include the image of other guide, such as animal painting, flowers and plants image, machine image etc..
Fig. 1 is a kind of flow chart of image base construction method provided in an embodiment of the present invention.
In S101, alignment operation is carried out to N width image, the N is the integer more than 1.
In embodiments of the present invention, alignment operation is carried out to N width image, can is by a certain same subject in N width images
Picture material carry out alignment operation.For example, when image is facial image, alignment operation is referred to carries out the face in image
Alignment operation so that size of the face in N width images in same level position and face is substantially consistent.Specifically, when
When image is facial image, alignment operation is entered to N width images includes as shown in Fig. 2 described：
Default face in the N width image are aligned to same level position by S201.
S202, adjusts the size of the N width image, until the size of the default face in the N width image is consistent.
In the corresponding embodiments of Fig. 2, the alignment of facial image can be realized on the basis of a certain face in face
Operation.Specifically, first N width image is carried out into horizontal position adjustment, so that certain face in this N width image are aligned to same water
Prosposition is put, and for example, the eyes for making face in N width images are respectively positioned on same level position, or the face for making face in N width images
It is respectively positioned on same level position.After horizontal position adjustment is completed, in addition it is also necessary to which the size of N width images is adjusted, with
Make the size of the face in this N width image consistent.For example, when default face are eyes, need to adjust the N width image
Size, until the binocular interval in the N width image is identical；Or, when default face are face, need to adjust N width images
Size, until the width of face is consistent in N width images；Or, when default face are nose, need to adjust N width figures
The size of picture, until the width of the wing of nose is consistent in N width images.By the corresponding embodiments of Fig. 2, facial image just can be realized
The alignment of middle human face region, to ensure N width images in face size it is consistent, it is to avoid cause because face is not of uniform size
The new facial image being subsequently generated do not meet face actual proportions situation occur.
Preferably, align image after by all image croppings be same size, for example, for facial image,
While ensureing the wherein integrity of face part, by N width image croppings into formed objects, so that the last image library set up
Middle image pattern more specification, it is easy to use.
In S102, to alignment after the N width image carry out predetermined patterns critical point detection respectively.
The predetermined patterns include：Eyebrow, eyes, nose, face, face mask etc..
Specifically, can by CFSS (Face Alignment by Coarse-to-Fine Shape Searching,
By thick to smart searching and detecting method), SDM (Supervised Descent Method, there is the gradient descent method of supervision) or
LBF (Local Binary Fitting, local binary fitting algorithm) scheduling algorithm carries out critical point detection.For example, when default portion
When being divided into eyebrow, by critical point detection, the key points such as the brows of eyebrow in image, eyebrow peak, eyebrow tail can be detected, based on this
A little key points, just can determine that eyebrow shared region in the picture.
In S103, according to the result of the predetermined patterns critical point detection, position in the N width image and extract pre-
If the area image at position.
Specifically, according to critical point detection result, the position such as positioning eyebrow, eyes, nose, face, face mask
Image-region, and by the image zooming-out in the image-region out.
In S104, the area image that identical predetermined patterns are carried out in the N width image is exchanged, and will be obtained
New images are stored in image library.
Specifically, as shown in figure 3, the area image that the identical predetermined patterns are carried out in the N width image is mutual
Change including：
S301, extracts the area image of the predetermined patterns in the wherein piece image of the N width image.
S302, by the area image of the predetermined patterns for extracting the corresponding position of remaining N-1 width image is superimposed to respectively
Put, to generate N-1 width new images.
In the corresponding embodiments of Fig. 3, it is accomplished that the administrative division map of the predetermined fraction of wherein piece image in N width images
As replacing to the same area of remaining N-1 width image, you can to generate N-1 width new images.Due to by between any two width image
Once the area image of identical predetermined patterns is exchanged can produce the new image of two width, then it is contemplated that in N width images,
The area image of identical predetermined patterns is exchanged between any two two image, or, further, in N width images, any two two
The area image of multiple predetermined patterns carries out area image exchange between image, then can produce substantial amounts of new images, so as to can
More new image patterns are generated simply and efficiently to pass through a small amount of image pattern, to complete the structure of image library.
Fig. 4 A-4E show the process of synthesis new images by taking two width images as an example, and the width image of the left side two is original image,
Two width images of the right are the image after synthesis, specially：
Fig. 4 A are the schematic diagrams before and after exchange face provided in an embodiment of the present invention；
Fig. 4 B are the schematic diagrams before and after exchange left eye provided in an embodiment of the present invention；
Fig. 4 C are the schematic diagrams before and after exchange right eye provided in an embodiment of the present invention；
Fig. 4 D are the schematic diagrams before and after exchange nose provided in an embodiment of the present invention；
Fig. 4 E are the schematic diagrams exchanged before and after remaining face provided in an embodiment of the present invention；
Wherein, in Fig. 4 E remaining face i.e. except the face after right and left eyes, face, nose.
Specifically, as shown in Figure 4 A,
Two original images are replicated respectively first：
Then the face area image of Ms (man) image is extracted；
By the face area image of Ms (man) image be added to replicate after man (Ms) image face region
On image, so as to form a new image.
In the same manner, the composograph process of Fig. 4 B-4E is as the process of Fig. 4 A.
Fig. 5 is the schematic diagram that 5 people provided in an embodiment of the present invention synthesize the partial synthesis design sketch in 100 images.
Fig. 6 shows the parts of images in the expression-A of third-party testing storehouse, the wherein image in here data base
Including sad expression and not sad expression, totally 752 images, two classes expression is each 376.
Table 1
| Method | Optimal test accuracy rate (%) | Speed (ms) |
| Original image | 51.46 | 30.01 |
| Oversampler method | 52.13 | 25.39 |
| Composograph | 59.97 | 25.45 |
Table 1 shows test result of the best model of training sad expression on third party library.As can be seen from the above table：
In terms of the sad expression identification of face, its accuracy rate is 59.97% to embodiment of the present invention synthetic method, is compared former
The best test result of beginning data training pattern is high by 8.51%, higher by 7.84% than by the best test result of oversampler method；
And embodiment of the present invention recognition speed is very fast, the fast 4.56ms of recognition speed of initial data training pattern is compared, than passing through to adopt
The fast 9.94ms of recognition speed of quadrat method.
So, the training sample that face character analysis is increased using embodiment of the present invention synthetic method is a kind of very good
Method.
A kind of method of the structure image library provided corresponding to the inventive embodiments, Fig. 7 shows that the embodiment of the present invention is carried
For a kind of image library construction device structured flowchart, for convenience of description, illustrate only part related to the present embodiment.
With reference to Fig. 7, the device includes：
Alignment unit 71, alignment operation is carried out to N width image, and the N is the integer more than 1；
Detector unit 72, to alignment after the N width image carry out predetermined patterns critical point detection respectively；
Extraction unit 73, according to the result of the predetermined patterns critical point detection, positions and extracts in the N width image
The area image of predetermined patterns；
Exchange unit 74, the area image that the identical predetermined patterns are carried out in the N width image is exchanged, and will be obtained
New images be stored in image library.
Alternatively, described image is facial image, described to enter alignment operation to N width images and include：
Default face in the N width image are aligned to into same level position；
The size of the N width image is adjusted, until the size of the default face in the N width image is consistent.
Alternatively, the default face include eyes, the size of the adjustment N width image, until the N width image
In default face size it is consistent including：
The size of the N width image is adjusted, until the binocular interval in the N width image is identical.
Alternatively, the N width image after described pair of alignment carries out respectively predetermined patterns critical point detection includes：
The N width image after following arbitrary algorithm is to alignment carries out respectively the critical point detection of predetermined patterns：By
Slightly to smart searching and detecting method CFSS, the gradient descent method SDM for having supervision and local binary fitting algorithm LBF.
Alternatively, the area image that the identical predetermined patterns are carried out in the N width image is exchanged and included：
The area image of the predetermined patterns is extracted in the wherein piece image of the N width image；
The area image of the predetermined patterns for extracting is superimposed to into respectively the correspondence position of remaining N-1 width image, with
Generate N-1 width new images.
In embodiments of the present invention, operated by image alignment, predetermined patterns critical point detection, extract the area of predetermined patterns
Area image, carries out area image exchange, so as to synthesize new image, based on a small amount of image pattern large sample amount just can be generated
Image library.
Presently preferred embodiments of the present invention is the foregoing is only, not to limit the present invention, all essences in the present invention
Any modification, equivalent and improvement made within god and principle etc., should be included within the scope of the present invention.
Claims (10)
1. a kind of image base construction method, it is characterised in that include：
Alignment operation is carried out to N width image, the N is the integer more than 1；
The N width image after to alignment carries out respectively the critical point detection of predetermined patterns；
According to the result of the critical point detection, the area image of the predetermined patterns is positioned and extracted in the N width image；
The area image that the identical predetermined patterns are carried out in the N width image is exchanged, and the new images for obtaining are stored in into figure
As storehouse.
2. the method for claim 1, it is characterised in that described image is facial image, it is described that alignment is entered to N width image
Operation includes：
Default face in the N width image are aligned to into same level position；
The size of the N width image is adjusted, until the size of the default face in the N width image is consistent.
3. method as claimed in claim 2, it is characterised in that the default face include eyes, the adjustment N width figure
The size of picture, until the size of the default face in the N width image it is consistent including：
The size of the N width image is adjusted, until the binocular interval in the N width image is identical.
4. the method for claim 1, it is characterised in that the N width image after described pair of alignment is preset respectively
The critical point detection at position includes：
The N width image after following arbitrary algorithm is to alignment carries out respectively the critical point detection of predetermined patterns：By slightly to
Searching and detecting method CFSS of essence, the gradient descent method SDM for having supervision and local binary fitting algorithm LBF.
5. the method for claim 1, it is characterised in that described that the identical default portion is carried out in the N width image
The area image of position is exchanged to be included：
The area image of the predetermined patterns is extracted in the wherein piece image of the N width image；
The area image of the predetermined patterns for extracting is superimposed to into respectively the correspondence position of remaining N-1 width image, to generate
N-1 width new images.
6. a kind of image library construction device, it is characterised in that include：
Alignment unit, for carrying out alignment operation to N width image, the N is the integer more than 1；
Detector unit, to alignment after the N width image carry out the critical point detection of predetermined patterns respectively；
Extraction unit, for according to the result of the critical point detection, in the N width image predetermined patterns being positioned and extracting
Area image；
Exchange unit, the area image for carrying out identical predetermined patterns in the N width image is exchanged, and will be obtained
New images are stored in image library.
7. device as claimed in claim 6, it is characterised in that described image is facial image, it is described that alignment is entered to N width image
Operation includes：
Default face in the N width image are aligned to into same level position；
The size of the N width image is adjusted, until the size of the default face in the N width image is consistent.
8. method as claimed in claim 7, it is characterised in that the default face include eyes, the adjustment N width figure
The size of picture, until the size of the default face in the N width image it is consistent including：
The size of the N width image is adjusted, until the binocular interval in the N width image is identical.
9. device as claimed in claim 6, it is characterised in that the N width image after described pair of alignment is preset respectively
Position critical point detection includes：
The N width image after following arbitrary algorithm is to alignment carries out respectively the critical point detection of predetermined patterns：By slightly to
Searching and detecting method CFSS of essence, the gradient descent method SDM for having supervision and local binary fitting algorithm LBF.
10. device as claimed in claim 6, it is characterised in that described that the identical default portion is carried out in the N width image
The area image of position is exchanged to be included：
The area image of the predetermined patterns is extracted in the wherein piece image of the N width image；
The area image of the predetermined patterns for extracting is superimposed to into respectively the correspondence position of remaining N-1 width image, to generate
N-1 width is newly schemed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611099286.9A CN106682587A (en) | 2016-12-02 | 2016-12-02 | Image database building method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611099286.9A CN106682587A (en) | 2016-12-02 | 2016-12-02 | Image database building method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106682587A true CN106682587A (en) | 2017-05-17 |
Family
ID=58867355
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611099286.9A Pending CN106682587A (en) | 2016-12-02 | 2016-12-02 | Image database building method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106682587A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111742342A (en) * | 2018-03-12 | 2020-10-02 | 日立产业控制解决方案有限公司 | Image generation method, image generation device, and image generation system |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101877146A (en) * | 2010-07-15 | 2010-11-03 | 北京工业大学 | Method for extending three-dimensional face database |
| CN105095857A (en) * | 2015-06-26 | 2015-11-25 | 上海交通大学 | Face data enhancement method based on key point disturbance technology |
-
2016
- 2016-12-02 CN CN201611099286.9A patent/CN106682587A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101877146A (en) * | 2010-07-15 | 2010-11-03 | 北京工业大学 | Method for extending three-dimensional face database |
| CN105095857A (en) * | 2015-06-26 | 2015-11-25 | 上海交通大学 | Face data enhancement method based on key point disturbance technology |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111742342A (en) * | 2018-03-12 | 2020-10-02 | 日立产业控制解决方案有限公司 | Image generation method, image generation device, and image generation system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10198623B2 (en) | Three-dimensional facial recognition method and system | |
| CN103914686B (en) | A kind of face alignment authentication method and system shone based on certificate photo with collection | |
| CN105426455B (en) | The method and apparatus that Classification Management is carried out to clothes based on picture processing | |
| CN108288027A (en) | A kind of detection method of picture quality, device and equipment | |
| CN106709936A (en) | Single target tracking method based on convolution neural network | |
| CN108090450B (en) | Face recognition method and device | |
| CN102567716B (en) | Face synthetic system and implementation method | |
| CN106778584A (en) | A kind of face age estimation method based on further feature Yu shallow-layer Fusion Features | |
| CN105404861A (en) | Training and detecting methods and systems for key human facial feature point detection model | |
| CN110110646A (en) | A kind of images of gestures extraction method of key frame based on deep learning | |
| CN104615996B (en) | A kind of various visual angles two-dimension human face automatic positioning method for characteristic point | |
| CN108108676A (en) | Face identification method, convolutional neural networks generation method and device | |
| CN109583443A (en) | A kind of video content judgment method based on Text region | |
| CN108520254A (en) | A kind of Method for text detection, device and relevant device based on formatted image | |
| CN106778852A (en) | A kind of picture material recognition methods for correcting erroneous judgement | |
| CN109598234A (en) | Critical point detection method and apparatus | |
| CN110852257B (en) | Method and device for detecting key points of human face and storage medium | |
| CN104778238B (en) | The analysis method and device of a kind of saliency | |
| CN105139007B (en) | Man face characteristic point positioning method and device | |
| CN110796101A (en) | Face recognition method and system of embedded platform | |
| CN107103311A (en) | A kind of recognition methods of continuous sign language and its device | |
| CN101493887A (en) | Eyebrow image segmentation method based on semi-supervision learning and Hash index | |
| CN110135230A (en) | Expression recognizes training system and expression recognizes training method | |
| CN106408086A (en) | Deep learning neural network processing method and deep learning neural network processing system for image optimization | |
| CN109284760A (en) | A kind of furniture detection method and device based on depth convolutional neural networks |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20170821 Address after: 361000, Xiamen three software park, Fujian Province, 8 North Street, room 2001 Applicant after: Xiamen Central Intelligent Information Technology Co., Ltd. Address before: 361000 Fujian province Xiamen software park two sunrise Road No. 32 403 unit 02 District Applicant before: XIAMEN ZHONGKONG BIOLOGICAL RECOGNITION INFORMATION TECHNOLOGY CO., LTD. |
|
| TA01 | Transfer of patent application right | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170517 |
|
| RJ01 | Rejection of invention patent application after publication |