CN102103695A - Method and device for generating image sample - Google Patents
Method and device for generating image sample Download PDFInfo
- Publication number
- CN102103695A CN102103695A CN2009102429913A CN200910242991A CN102103695A CN 102103695 A CN102103695 A CN 102103695A CN 2009102429913 A CN2009102429913 A CN 2009102429913A CN 200910242991 A CN200910242991 A CN 200910242991A CN 102103695 A CN102103695 A CN 102103695A
- Authority
- CN
- China
- Prior art keywords
- image pattern
- aam
- image
- sample
- image sample
- 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
- 238000000034 method Methods 0.000 title claims abstract description 21
- 238000012549 training Methods 0.000 claims abstract description 26
- 230000001815 facial effect Effects 0.000 claims description 58
- 239000013598 vector Substances 0.000 claims description 13
- 238000006243 chemical reaction Methods 0.000 claims description 10
- 230000014509 gene expression Effects 0.000 claims description 7
- 230000009466 transformation Effects 0.000 claims description 7
- 238000000513 principal component analysis Methods 0.000 description 12
- 238000005516 engineering process Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 description 2
- 238000013179 statistical model Methods 0.000 description 2
- 238000013480 data collection Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
Images
Landscapes
- Image Processing (AREA)
Abstract
The invention discloses a method and device for generating image sample for obtaining the image sample featured with better quality and excellent representativeness for expanding an image sample base, improving the training precision of a classifier of the image sample and satisfying the application demand. The invention provides a method for generating image sample, comprising the following steps of: classifying the collected image samples according to the predetermined type to obtain different types of image sample sets; performing AAM (Active Appearance Model) training to the image sample of each image sample set to obtain a plurality of AAM models; and generating new image sample by each AAM model.
Description
Technical field
The present invention relates to technical field of image processing, relate in particular to a kind of image pattern generation method and device.
Background technology
In the technology that relates to the facial image processing is used, detect as people's face, recognitions of face etc., the collection of facial image sample image and processing are very crucial problems.
Prior art generally is to collect a large amount of facial image sample training, and sample is many more, and the sorter that training obtains just has extendability more.But the collection of a large amount of facial image samples is wasted time and energy, and the sample of collecting is irregular not strange yet.Especially some special applications, when for example Real time identification was single, the facial image sample was limited.
In order to obtain how new facial image sample according to existing facial image sample, prior art has proposed certain methods, for example: utilize the mirror transformation technology to obtain the image pattern of left and right sides face symmetry, perhaps utilize a plurality of facial images to calculate an average facial image, as new facial image sample etc.
But the method that the above-mentioned prior art that these obtain new facial image sample adopts is fairly simple, and new less, the limited amount of facial image specimen types of generation often can't satisfy the needs of practical application.
Summary of the invention
The embodiment of the invention provides a kind of image pattern generation method and device, in order to obtain better quality, how representative image pattern, thereby enlarge the image pattern storehouse, improve training precision, satisfy more application demands based on the sorter of image pattern.
A kind of image pattern generation method that the embodiment of the invention provides comprises:
According to the classification that sets in advance, the image pattern that collects is classified, obtain various different classes of image pattern set;
Respectively the image pattern in each image pattern set is carried out the AAM training, obtain a plurality of AAM models;
Utilize each AAM model to generate new image pattern.
A kind of image pattern generating apparatus that the embodiment of the invention provides comprises:
Image pattern set generation unit is used for according to the classification that sets in advance the image pattern that collects being classified, and obtains various different classes of image pattern set;
AAM model generation unit, the image pattern that is used for respectively each image pattern being gathered carry out the AAM training, obtain a plurality of AAM models;
New images sample generation unit is used to utilize each AAM model to generate new image pattern.
The embodiment of the invention is classified to the image pattern that collects according to the classification that sets in advance, and obtains various different classes of image pattern set; Respectively the image pattern in each image pattern set is carried out the AAM training, obtain a plurality of AAM models; Utilize each AAM model to generate new image pattern, thereby can obtain better quality, how representative image pattern, enlarged the image pattern storehouse, improved training precision, satisfied the more applications demand based on the sorter of image pattern.
Description of drawings
The overall procedure synoptic diagram of a kind of image pattern generation method that Fig. 1 provides for the embodiment of the invention;
The generating principle synoptic diagram of the facial image sample that Fig. 2 provides for the embodiment of the invention;
The variation synoptic diagram of the movable contour model that Fig. 3 provides for the embodiment of the invention (ASM, Active Shape Model) model-driven shape;
The generation synoptic diagram of the sample that is used for texture model training that Fig. 4 provides for the embodiment of the invention;
Fig. 5 generates the synoptic diagram of different facial image models for the movable appearance model (AAM, Active AppearanceModel) that utilizes that the embodiment of the invention provides;
The structural representation of a kind of image pattern generating apparatus that Fig. 6 provides for the embodiment of the invention.
Embodiment
The embodiment of the invention provides a kind of image pattern generation method and device, in order to obtain better quality, how representative image pattern, thereby enlarge the image pattern storehouse, improve training precision, satisfy more application demands based on the sorter of image pattern.
The embodiment of the invention is that example describes with the facial image sample.Certainly, for the image pattern of other objects, the technical scheme that the embodiment of the invention provides is suitable equally.
The embodiment of the invention generates new facial image sample based on the AAM model, utilize the AAM model from limited facial image sample, to generate various new facial image samples, it is fine to obtain quality, simultaneously representative facial image sample, and can generate infinite a plurality of new facial image sample, thereby enlarge facial image sample storehouse, improve training precision based on the sorter of facial image.
In the application (as people's face location, Expression Recognition) based on a large amount of facial image samples, the quality and quantity of facial image sample is very big to the net result influence of training.The facial image sample size is many and do not mean that the facial image sample can cover various people's face situations.In the facial image sample, may exist many similar facial image samples, the embodiment of the invention is based on the AAM technology, utilize facial image sample storehouse to train AAM facial image model, principal component after the utilization training is controlled the variation of facial image then, can obtain various new person's face image patterns of having represented different light, different expressions etc.The a large amount of facial image sample information of these new facial image sample utilizations is synthesized, and therefore has very strong representativeness.
In addition, utilize the AAM model can also generate many extreme expressions and extreme illumination condition facial image sample down, these extreme facial image samples generally are difficult to collect, so can enlarge the representativeness in facial image sample storehouse.
In the embodiment of the invention,, utilize AAM to generate new facial image sample then, everyone facial image sample is expanded the image pattern of everyone face training AAM model.For example, utilize the image pattern of others face of homogeny to train the AAM model, can obtain other new facial image sample of homogeny; The image pattern of people's face of all ages and classes section is carried out the AAM training, can obtain the new facial image sample of all ages and classes section.
Below in conjunction with accompanying drawing the technical scheme that the embodiment of the invention provides is described.
Referring to Fig. 1, a kind of image pattern generation method that the embodiment of the invention provides totally comprises step:
S101, according to the classification that sets in advance, the image pattern that collects is classified, obtain the set of various different classes of image patterns.
S102, respectively the image pattern in the set of each image pattern is carried out the AAM training, obtain a plurality of AAM models.
S103, utilize each AAM model to generate new image pattern.
Preferably, step S103 comprises:
To each AAM model, will make up by eigenwert and the proper vector that principal component analysis (PCA) (PCA) conversion obtains, generate new image pattern.
PCA is a kind of technology of reduced data collection.It is a linear transformation.This conversion transforms the data in the new coordinate system, makes the first variance of any data projection on first coordinate (being called first principal component), second largest variance on second coordinate (Second principal component), and the like.Principal component analysis (PCA) keeps the feature to the variance contribution maximum of data set simultaneously through the dimension that reduces data set commonly used.This is by keeping the low order major component, ignores that the high-order major component accomplishes.The low order composition often can retain the most important aspect of data like this.Major component is a proper vector, the major component that variance is big, and its characteristic of correspondence value is just big more.If for each composition distributes a weight, be weighted then and add up, just can make up and obtain new data.
Preferably, this method also comprises:
New image pattern is carried out mirror transformation, obtain the mirror image sample.
Preferably, described image pattern set comprises:
The facial image sample set of same people's face under different light;
The facial image sample set of same people's face under the difference expression;
Others gathers homogeneity by the face image pattern;
Be distributed in the facial image sample set of all ages and classes section.
Below the AAM model described in the embodiment of the invention is described below:
AAM is a kind of powerful tool of image understanding.It is based upon on the statistical basis, has considered the texture information in body form and the shape overlay area simultaneously.AAM comprises two parts, and a part is that shape is added up, and a part is the statistics to the texture information of shape inside in addition.
The model of separately shape being added up is the ASM model, and the ASM model is a kind of expression that the people's face shape to the sample space representative changes, and therefore can be used for controlling the variation of the facial image shape of different attitudes in the human face locations of contours.
What AAM reflected is the variation of people's face gray scale, it forms a sample vector with each pixel of people's face picture of prescribed level according to a series arrangement, then a large amount of gray scale sample vectors is carried out the PCA conversion, just can obtain reflecting the principal component of different variation characteristics.Different principal components is made up according to different weights, just can obtain infinite a plurality of facial image.
Training is described below to the AAM described in the embodiment of the invention below:
Training and the ASM of AAM are similar.It is one one n dimensional vector n that set of coordinates that at first will manual point of demarcating is made into.If the coordinate of the facial contour point in facial image is: (x
i, y
i), wherein, i=0,1 ..., n, n are the number of point.
So, the planimetric coordinates tissue of all point on this facial image can be become one one n dimensional vector n:
S
j=[x
0,y
0,...,x
n,y
n]
S then
jIt is exactly a sample of training.
The facial contour point of all demarcation on this facial image all is organized into a n dimensional vector n, then all samples is carried out the PCA conversion and just can obtain the ASM model:
Wherein, s
0Be the average of training sample, s
iBe the characteristic component consistent that the PCA conversion obtains with sample size, s
iCan be used for describing the variation of facial image shape, as shown in Figure 3.
The sample collection of people's face texture variations statistical model is more complicated than ASM.At first need to determine mean profile face s
0Size, the gray-scale value of the pixel in the zone that covers with average face is as the element of vector sample.The mean profile face can obtain according to ASM.In order to obtain the gray-scale value of the pixel in the average face overlay area, at first to the point structure key triangulation network, the coordinate of the pixel after then each triangle in the triangulation network being carried out discretize and writes down discretize.
As shown in Figure 4, to any one people's face s that has demarcated point, set up s according to the topological result of mean profile triangle gridding
0And the affine relation between each triangle between the s.This affine relation can be used W (X; P) represent that X represents s
0The coordinate of a last point, the p correspondence parameter of the affined transformation between every diabolo.According to W (X; P) just can from actual facial image, sample out and the little duplicate zone of on average being bold.
The veining structure that mean profile is comprised each pixel in the scope is a vector sample, and all vector samples are carried out the statistical model that the PCA conversion just can obtain texture information:
The same with ASM, A
0(X) be the average of all samples, A
i(X) be the characteristic component that obtains with the PCA conversion.Change each characteristic component corresponding parameters λ
i, can obtain different textures, effect is as shown in Figure 5.
Referring to Fig. 6, a kind of image pattern generating apparatus that the embodiment of the invention provides comprises:
Image pattern set generation unit 11 is used for according to the classification that sets in advance the image pattern that collects being classified, and obtains various different classes of image pattern set.
AAM model generation unit 12, the image pattern that is used for respectively each image pattern being gathered carry out the AAM training, obtain a plurality of AAM models.
New images sample generation unit 13 is used to utilize each AAM model to generate new image pattern.
Preferably, described new images sample generation unit 13 to each AAM model, will make up by eigenwert and the proper vector that the PCA conversion obtains, and generates new image pattern.
Preferably, this device also comprises:
Mirror image sample generation unit 14 is used for new image pattern is carried out mirror transformation, obtains the mirror image sample.
In sum, the embodiment of the invention is classified to the image pattern that collects according to the classification that sets in advance, and obtains various different classes of image pattern set; Respectively the image pattern in each image pattern set is carried out the AAM training, obtain a plurality of AAM models; Utilize each AAM model to generate new image pattern, thereby can obtain better quality, how representative image pattern, enlarged the image pattern storehouse, improved training precision, satisfied the more applications demand based on the sorter of image pattern.
Obviously, those skilled in the art can carry out various changes and modification to the present invention and not break away from the spirit and scope of the present invention.Like this, if of the present invention these are revised and modification belongs within the scope of claim of the present invention and equivalent technologies thereof, then the present invention also is intended to comprise these changes and modification interior.
Claims (10)
1. an image pattern generation method is characterized in that, this method comprises:
According to the classification that sets in advance, the image pattern that collects is classified, obtain various different classes of image pattern set;
Respectively the image pattern in each image pattern set is carried out the AAM training, obtain a plurality of AAM models;
Utilize each AAM model to generate new image pattern.
2. method according to claim 1 is characterized in that, the step of utilizing each AAM model to generate new image pattern comprises:
To each AAM model, will make up by eigenwert and the proper vector that the PCA conversion obtains, generate new image pattern.
3. method according to claim 1 is characterized in that, this method also comprises:
Described new image pattern is carried out mirror transformation, obtain the mirror image sample.
4. according to claim 1,2 or 3 described methods, it is characterized in that described image pattern is people's face image pattern.
5. method according to claim 4 is characterized in that, described image pattern set comprises:
The facial image sample set of same people's face under different light;
The facial image sample set of same people's face under the difference expression;
Others gathers homogeneity by the face image pattern;
Be distributed in the facial image sample set of all ages and classes section.
6. an image pattern generating apparatus is characterized in that, this device comprises:
Image pattern set generation unit is used for according to the classification that sets in advance the image pattern that collects being classified, and obtains various different classes of image pattern set;
AAM model generation unit, the image pattern that is used for respectively each image pattern being gathered carry out the AAM training, obtain a plurality of AAM models;
New images sample generation unit is used to utilize each AAM model to generate new image pattern.
7. device according to claim 6 is characterized in that, described new images sample generation unit to each AAM model, will make up by eigenwert and the proper vector that the PCA conversion obtains, and generates new image pattern.
8. device according to claim 6 is characterized in that, this device also comprises:
Mirror image sample generation unit is used for described new image pattern is carried out mirror transformation, obtains the mirror image sample.
9. according to claim 6,7 or 8 described devices, it is characterized in that described image pattern is people's face image pattern.
10. device according to claim 6 is characterized in that, the image pattern set that described image pattern set generation unit generates comprises:
The facial image sample set of same people's face under different light;
The facial image sample set of same people's face under the difference expression;
Others gathers homogeneity by the face image pattern;
Be distributed in the facial image sample set of all ages and classes section.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2009102429913A CN102103695A (en) | 2009-12-21 | 2009-12-21 | Method and device for generating image sample |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2009102429913A CN102103695A (en) | 2009-12-21 | 2009-12-21 | Method and device for generating image sample |
Publications (1)
Publication Number | Publication Date |
---|---|
CN102103695A true CN102103695A (en) | 2011-06-22 |
Family
ID=44156450
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN2009102429913A Pending CN102103695A (en) | 2009-12-21 | 2009-12-21 | Method and device for generating image sample |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN102103695A (en) |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104112131A (en) * | 2013-04-19 | 2014-10-22 | 浙江大华技术股份有限公司 | Method and device for generating training samples used for face detection |
CN104700076A (en) * | 2015-02-13 | 2015-06-10 | 电子科技大学 | Face image virtual sample generating method |
CN104715224A (en) * | 2013-12-11 | 2015-06-17 | 腾讯科技(深圳)有限公司 | Method and device for acquiring facial feature images of user group |
CN108230233A (en) * | 2017-05-16 | 2018-06-29 | 北京市商汤科技开发有限公司 | Data enhancing, treating method and apparatus, electronic equipment and computer storage media |
CN109063748A (en) * | 2018-07-16 | 2018-12-21 | 重庆大学 | Object detection method based on data enhancing |
CN109583302A (en) * | 2018-10-29 | 2019-04-05 | 杭州电子科技大学 | A kind of mobile robot area of feasible solutions training dataset extending method |
CN110147845A (en) * | 2019-05-23 | 2019-08-20 | 北京百度网讯科技有限公司 | Sample collection method and sample acquisition system based on feature space |
CN110689087A (en) * | 2019-10-10 | 2020-01-14 | 西南石油大学 | Image sample generation method based on probability likelihood |
CN110717453A (en) * | 2019-10-10 | 2020-01-21 | 西南石油大学 | Image sample generation method based on cosine distance |
CN110877019A (en) * | 2018-09-05 | 2020-03-13 | 西门子(中国)有限公司 | Traditional Chinese medicinal material impurity removing device and method |
CN111079624A (en) * | 2019-12-11 | 2020-04-28 | 北京金山云网络技术有限公司 | Method, device, electronic equipment and medium for collecting sample information |
CN111160529A (en) * | 2019-12-28 | 2020-05-15 | 天津大学 | Convolutional neural network-based training sample generation method in target pose measurement |
CN111178337A (en) * | 2020-01-07 | 2020-05-19 | 南京甄视智能科技有限公司 | Human face key point data enhancement method, device and system and model training method |
WO2021217856A1 (en) * | 2020-04-30 | 2021-11-04 | 平安科技(深圳)有限公司 | Face image generation method and apparatus, electronic device, and readable storage medium |
-
2009
- 2009-12-21 CN CN2009102429913A patent/CN102103695A/en active Pending
Cited By (24)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104112131A (en) * | 2013-04-19 | 2014-10-22 | 浙江大华技术股份有限公司 | Method and device for generating training samples used for face detection |
CN104112131B (en) * | 2013-04-19 | 2017-03-22 | 浙江大华技术股份有限公司 | Method and device for generating training samples used for face detection |
CN104715224A (en) * | 2013-12-11 | 2015-06-17 | 腾讯科技(深圳)有限公司 | Method and device for acquiring facial feature images of user group |
CN104715224B (en) * | 2013-12-11 | 2019-01-04 | 腾讯科技(深圳)有限公司 | A kind of method and device for the facial feature image obtaining user group |
CN104700076A (en) * | 2015-02-13 | 2015-06-10 | 电子科技大学 | Face image virtual sample generating method |
CN104700076B (en) * | 2015-02-13 | 2017-09-12 | 电子科技大学 | Facial image virtual sample generation method |
CN108230233A (en) * | 2017-05-16 | 2018-06-29 | 北京市商汤科技开发有限公司 | Data enhancing, treating method and apparatus, electronic equipment and computer storage media |
CN109063748A (en) * | 2018-07-16 | 2018-12-21 | 重庆大学 | Object detection method based on data enhancing |
CN109063748B (en) * | 2018-07-16 | 2021-11-16 | 重庆大学 | Target detection method based on data enhancement |
CN110877019A (en) * | 2018-09-05 | 2020-03-13 | 西门子(中国)有限公司 | Traditional Chinese medicinal material impurity removing device and method |
CN109583302A (en) * | 2018-10-29 | 2019-04-05 | 杭州电子科技大学 | A kind of mobile robot area of feasible solutions training dataset extending method |
CN109583302B (en) * | 2018-10-29 | 2020-09-18 | 杭州电子科技大学 | Mobile robot feasible area training data set expansion method |
CN110147845B (en) * | 2019-05-23 | 2021-08-06 | 北京百度网讯科技有限公司 | Sample collection method and sample collection system based on feature space |
CN110147845A (en) * | 2019-05-23 | 2019-08-20 | 北京百度网讯科技有限公司 | Sample collection method and sample acquisition system based on feature space |
CN110717453A (en) * | 2019-10-10 | 2020-01-21 | 西南石油大学 | Image sample generation method based on cosine distance |
CN110689087A (en) * | 2019-10-10 | 2020-01-14 | 西南石油大学 | Image sample generation method based on probability likelihood |
CN110689087B (en) * | 2019-10-10 | 2023-04-18 | 西南石油大学 | Image sample generation method based on probability likelihood |
CN111079624A (en) * | 2019-12-11 | 2020-04-28 | 北京金山云网络技术有限公司 | Method, device, electronic equipment and medium for collecting sample information |
CN111079624B (en) * | 2019-12-11 | 2023-09-01 | 北京金山云网络技术有限公司 | Sample information acquisition method and device, electronic equipment and medium |
CN111160529A (en) * | 2019-12-28 | 2020-05-15 | 天津大学 | Convolutional neural network-based training sample generation method in target pose measurement |
CN111160529B (en) * | 2019-12-28 | 2023-06-20 | 天津大学 | Training sample generation method in target pose measurement based on convolutional neural network |
CN111178337B (en) * | 2020-01-07 | 2020-12-29 | 南京甄视智能科技有限公司 | Human face key point data enhancement method, device and system and model training method |
CN111178337A (en) * | 2020-01-07 | 2020-05-19 | 南京甄视智能科技有限公司 | Human face key point data enhancement method, device and system and model training method |
WO2021217856A1 (en) * | 2020-04-30 | 2021-11-04 | 平安科技(深圳)有限公司 | Face image generation method and apparatus, electronic device, and readable storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN102103695A (en) | Method and device for generating image sample | |
CN109325437B (en) | Image processing method, device and system | |
KR101555347B1 (en) | Apparatus and method for generating video-guided facial animation | |
CN107330437B (en) | Feature extraction method based on convolutional neural network target real-time detection model | |
JP2016218999A (en) | Method for training classifier to detect object represented in image of target environment | |
CN109508360A (en) | A kind of polynary flow data space-time autocorrelation analysis method of geography based on cellular automata | |
CN108009509A (en) | Vehicle target detection method | |
CN109697434A (en) | A kind of Activity recognition method, apparatus and storage medium | |
CN106909875A (en) | Face shape of face sorting technique and system | |
CN110930387A (en) | Fabric defect detection method based on depth separable convolutional neural network | |
CN101276420A (en) | Classification method for syncretizing optical spectrum information and multi-point simulation space information | |
CN108154104A (en) | A kind of estimation method of human posture based on depth image super-pixel union feature | |
CN110163213A (en) | Remote sensing image segmentation method based on disparity map and multiple dimensioned depth network model | |
CN104036284A (en) | Adaboost algorithm based multi-scale pedestrian detection method | |
CN102509104A (en) | Confidence map-based method for distinguishing and detecting virtual object of augmented reality scene | |
CN114241422A (en) | Student classroom behavior detection method based on ESRGAN and improved YOLOv5s | |
CN109871829A (en) | A kind of detection model training method and device based on deep learning | |
WO2013055914A2 (en) | Systems and methods for creating texture exemplars | |
JP2014211719A (en) | Apparatus and method for information processing | |
CN107944437A (en) | A kind of Face detection method based on neutral net and integral image | |
CN109117717A (en) | A kind of city pedestrian detection method | |
CN109670401A (en) | A kind of action identification method based on skeleton motion figure | |
JP2010211732A (en) | Object recognition device and method | |
CN109117791A (en) | A kind of crowd density drawing generating method based on expansion convolution | |
CN108805022A (en) | A kind of remote sensing scene classification method based on multiple dimensioned CENTRIST features |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C12 | Rejection of a patent application after its publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20110622 |