CN108564095A - A kind of image texture sorting technique based on contrast local binary patterns - Google Patents
A kind of image texture sorting technique based on contrast local binary patterns Download PDFInfo
- Publication number
- CN108564095A CN108564095A CN201810379373.2A CN201810379373A CN108564095A CN 108564095 A CN108564095 A CN 108564095A CN 201810379373 A CN201810379373 A CN 201810379373A CN 108564095 A CN108564095 A CN 108564095A
- Authority
- CN
- China
- Prior art keywords
- feature
- image
- texture
- pixel
- extraction
- 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
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
- G06V10/507—Summing image-intensity values; Histogram projection analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification
Abstract
A kind of image texture sorting technique based on contrast local binary patterns, includes the following steps:Extract the symbolic feature of image;Extract the contrast difference energy feature of image;Extract the center pixel feature of image;The symbolic feature of the image of extraction, contrast difference energy feature and center pixel feature are integrated, the symbol energy center pixel histogram feature of image is obtained, and establishes histogram;Classified to histogram using chi-Square measure and nearest neighbor classifier.Advantageous effect of the present invention:The present invention improves the accuracy of feature extraction, classification.
Description
Technical field
The present invention relates to pattern-recognitions and technical field of image processing, specifically a kind of to be based on contrast local binary
The image texture sorting technique of pattern.
Background technology
With the development of the fast development and computer vision, pattern-recognition related field of computer technology so that image
Treatment technology is more and more ripe.Such as the photo search service that Google and Baidu have released.Baidu sets up Baidu's number
According to visualized experiment room, deep learning platform is made.We can carry out Baidu's knowledge figure by scanned picture;It is powerful in Taobao
View system so that input one commodity picture almost can accurately search identical or similar product.Such as
Face recognition in present iphoneX.Above-mentioned technology is all to find accurate figure later by carrying out feature extraction to image
Piece information.
Texture plays critically important role in Pattern recognition and image processing.Therefore, the feature extraction of texture is line
The core for managing research, is the basis of texture analysis and application.In recent years, feature extraction and Texture classification are carried out by large quantities of scholars
Research and extension, and it is widely used in recognition of face, medical biotechnology analysis, the fields such as car plate detection.Look back more than 50 years complications
Development course, various countries researcher conduct extensive research texture characteristic extracting method.
For a complete Texture classification usually by image preprocessing, feature extraction, class test these step groups
At.Wherein the most core also most influence experiment classification accuracy rate be exactly feature extracting method research.In computer graphics
With the field of pattern-recognition, the analysis of image texture is carried out like a raging fire.Nowadays, scholars have been proposed very much
Texture method.Two classes can be substantially divided into:Spatial domain method and transform domain method, wherein spatial domain method are divided into as Structure Method, statistics
This three classes of method, modelling.Spatial domain method.The method of this major class is mainly adjacent with it by comparing the center pixel of texture image
Relationship between the pixel of domain simultaneously carries out the texture analysis to image by statistic law to indicate the relationship of its texture.Popular comes
Say to be exactly the primitive of analog texture to reconfigure the space partial structurtes of image.This major class method, which is primarily adapted for use in, to be easy to revolve
The texture image turned.Transform domain typically uses transformation coefficient to analyze texture image.
The LBP algorithms of local shape factor only only account for center pixel and neighborhood when carrying out feature extraction to image
Symbolic information between pixel, the information of the gray value size without counting it.Meanwhile LBP algorithms have rotational invariance
While be lost directional information, many times directional information is very important a texture image, but in texture
In image analysis, LBP is proved to be effective.Local shape factor can find out that this method is extracted from the literal meaning
Be local feature, its method calculate it is simple, have and promote well.But it also has ignored the image texture of global macroscopic view
Information.Wherein, local binary patterns are widely discussed and are studied by everybody as one of classic algorithm.Its algorithm is simple,
It should be readily appreciated that, dimension is relatively low, calculating is simple and high resolution makes lot of domestic and international scientific research scholar all carry being engaged in local feature
The work taken.
Invention content
The texture of technical problem to be solved by the invention is to provide a kind of contrast local binary patterns based on extension
Sorting technique, its symbolic feature is only accounted in texture feature extraction for solution local binary patterns (LBP) and discrimination is relatively low
The problem of.
Used technical solution is the present invention to solve above-mentioned technical problem:One kind being based on contrast local binary patterns
Image texture sorting technique, include the following steps:
The symbolic feature of step 1, extraction texture image f (x, y);
The contrast difference energy feature of step 2, extraction texture image f (x, y);
The center pixel feature of step 3, extraction texture image f (x, y);
Step 4, symbolic feature, contrast difference energy feature and center pixel feature by the texture image f (x, y) of extraction
It is integrated, obtains the symbol energy center pixel histogram feature of texture image f (x, y), and establish histogram;
Step 5 classifies to histogram using chi-Square measure and nearest neighbor classifier.
The extraction of the symbolic feature of texture image f (x, y) is carried using local binary patterns in step 1 of the present invention
Take description.
The specific extracting method of the symbolic feature of texture image f (x, y) is in step 1 of the present invention:
Step 1.1, the module for defining a 3*3 select pixel centered on a pixel, compare in this on that module
Imago element and 8 neighborhood territory pixel g of surroundingPBetween size, the use 0 less than center pixel indicates, is greater than or equal to center pixel
Use 1 indicate that then the symbolic feature of texture image f (x, y) can be expressed as:
The symbolic feature of texture image is considered as end to end annular by step 1.2 according to a fixed direction, is obtained
The equivalent formulations of LBP, with 0 to 1 transition times U (GP) indicate the equivalent formulations, then:
U (the G of step 13, the equivalent formulations for obtaining step 1.2P) rotated to obtain and rotate not equivalent formulations of property
LBP can be specifically expressed as:
The symbolic feature histogram H of step 1.4, texture image f (x, y)iIt is represented by:
The specific method of the contrast difference energy feature of extraction texture image f (x, y) is in step 2 of the present invention:
Step 2.1, the contrast difference energy feature CDEF for calculating each pixel:It is fixed on texture image f (x, y)
The module of one 3*3 of justice selects pixel g centered on a pixel on that modulec, contrast difference energy feature CDEF can
It is calculated by following formula:Wherein gPFor gcNeighborhood territory pixel, 1≤P≤9;
Step 2.2, setting difference value FD:Wherein max is the maximum value of image pixel, and min is figure
As the minimum value of pixel;
Step 2.3, the section CV that bin histogram is set:13 critical value CV are set0,CV1..., CV12, value by
Following formula calculates:
CVi=H (x)+min i ∈ [0,12]
The specific method of the center pixel feature F (x, y) of extraction texture image f (x, y) is in step 3 of the present invention:F
(x, y)=c (x, y) mod n, n > 2, the value of the n is preferably 12.
The beneficial effects of the invention are as follows:The present invention not only increases the accuracy of feature extraction, classification.It also proposed simultaneously
The energy information of picture is extracted by contrast, it is proposed that the method for the big minizone of a new partition window improves special
Levy the accuracy of extraction.The binaryzation of script is changed into parameter by experiment simultaneously, intermediate processing is carried out to picture, carrys out table
The feature of diagram piece, more accurately.Experimental result on data set shows that sorting technique of the present invention has good classification
Can, and the calculating speed of this feature extracting method is quickly, and the dimension of feature is relatively small, energy while ensureing classification performance
Enough realize quick texture image classification.
Description of the drawings
Fig. 1 is the flow chart of image texture sorting technique of the present invention;
Fig. 2 is that symbolic feature of the present invention extracts mode method;
Fig. 3 is center pixel feature extracting method of the present invention;
Fig. 4 is the classification results that multiple sorting techniques are compared;
Fig. 5 is a class postrotational texture image on eight directions.
Specific implementation mode
A kind of image texture sorting technique based on contrast local binary patterns, includes the following steps:
The symbolic feature of step 1, extraction texture image f (x, y);
The contrast difference energy feature of step 2, extraction texture image f (x, y);
The center pixel feature of step 3, extraction texture image f (x, y);
Step 4, symbolic feature, contrast difference energy feature and center pixel feature by the texture image f (x, y) of extraction
It is integrated, obtains the symbol energy center pixel histogram feature of texture image f (x, y), and establish histogram;
Step 5 classifies to histogram using chi-Square measure and k nearest neighbor grader, and K preferably takes 1.
Further, the extraction of the symbolic feature of texture image f (x, y) is carried out using local binary patterns in the step 1
Extraction description.
Further, the specific extracting method of the symbolic feature of texture image f (x, y) is in the step 1:
Step 1.1, the module for defining a 3*3 select pixel centered on a pixel, compare in this on that module
Imago element and 8 neighborhood territory pixel g of surroundingPBetween size, the use 0 less than center pixel indicates, is greater than or equal to center pixel
Use 1 indicate that then the symbolic feature of image can be expressed as:
The symbolic feature of image is considered as end to end annular by step 1.2 according to a fixed direction, obtains LBP
Equivalent formulations, with 0 to 1 transition times U (GP) indicate the equivalent formulations, then:
U (the G of step 1.3, the equivalent formulations for obtaining step 1.2P) rotated to obtain the equivalent formulations for rotating not property
LBP, can specifically be expressed as:
The symbolic feature histogram H of step 1.4, texture image f (x, y)iIt is represented by:
Further, the specific method of the contrast difference energy feature of extraction texture image f (x, y) is in the step 2:
Step 2.1, the contrast difference energy feature CDEF for calculating each pixel:It is fixed on texture image f (x, y)
The module of one 3*3 of justice selects pixel g centered on a pixel on that modulec, contrast difference energy feature CDEF can
It is calculated by following formula:Wherein gPFor gcNeighborhood territory pixel, 1≤P≤9;
Step 2.2, setting difference value FD:When feature histogram is established, suitable bins is selected, it appears particularly important.
In view of the principle of contrast difference energy operator is to seek the difference of pixel and its field, and these differences are relatively small.Therefore
It is equidistant we used mode ascending bins in contrast difference energy feature when establishing histogram
It establishes feature histogram, sets a difference value FD,Wherein max is the maximum value of image pixel,
Min is the minimum value of image pixel;
Step 2.3, the section CV that bin histogram is set:13 critical value CV are set0,CV1..., CV12, value by
Following formula calculates:
CVi=H (x)+min i ∈ [0,12]
Further, the specific method of the center pixel feature F (x, y) of extraction texture image f (x, y) is in the step 3:
F (x, y)=f (x, y) modn, wherein parameter n > 2.When extracting the center pixel feature of f (x, y), we carry out image more
Value is handled, rather than binaryzation.The binaryzation of script is changed into parameter by experiment, intermediate processing is carried out to picture,
To indicate the feature of picture.Show that parameter takes 12 according to experiment, experiment effect is optimal.
Average correct classification rate (ACAR, Average Classification Accuracy Rate) as final
Evaluation criteria.Experimental result on the Brodatz of standard texture library is as shown in figure 4, Fig. 4 gives multiple comparative approach simultaneously
Classification results.
Embodiment
The detailed process of the present invention is illustrated by carrying out classified instance to the texture image in the Brodatz of standard texture library such as
Under:
(1) texture image being pre-processed, Brodatz texture searchings are an international texture image data libraries,
111 texture classes are shared, each texture classes is the gray scale texture image of a 640*640.It is pre-processed to texture collection
During, in our experiment, we are using the central area of each image of 320*320 as experimental image.In order to test
The rotational invariance of this method is demonstrate,proved, we rotate each 320*320 gray scales texture image.We are in the direction of the clock
Each class is rotated:0 degree, 5 degree, 10 degree, 30 degree, 45 degree, 60 degree, 75 degree, 90 degree, as shown in Figure 5.For the line
Reason collection, we devise three data sets, we take each postrotational 320*320 gray scales texture image class to be divided into 4
160*160 experiment sample, therefore each class corresponds to the experiment sample of 32 160*160, a half-sample of each class is random
Selection is for training, remaining is for testing.We have randomly selected three data sets, are respectively:Set-1, Set-2 and
Set-3, these three data sets separately include 960,1920 and 3552 experiment samples, as shown in Figure 5.
(2) the center pixel feature for calculating symbolic feature, contrast difference energy feature and extension, by symbolic feature, right
It is integrated than degree differential power feature and the center pixel feature of extension.To the partial structurtes Model Establishment of entire input picture
Statistic histogram is as neighborhood difference modes feature.
(3) classified to texture image with k nearest neighbor grader, obtain nicety of grading.
(4) we test method proposed by the invention with competing to LBP, CLBP, LRS-MD and PMC-BC respectively
Power is striven, is more than 3% better than CLBP_SMC and PMC-BC, while it is also seen that on whole Brodatz textures collection of Set-3 originally
The method of invention is also superior to LRS-MD and PMC-BC, as shown in Figure 4.By the comparison with other methods, the present invention can be verified
The method proposed has good superiority relative to other four algorithms:It is effectively improved the classification essence of texture image
Degree, effectively adapts to the variation of texture image image-forming condition.In addition, method proposed by the invention is relative to traditional local binary
The information of conventional method loss is adequately utilized on the basis of retaining calculating simply in pattern (LBP) method, has extensive
Application value.
Claims (6)
1. a kind of image texture sorting technique based on contrast local binary patterns, it is characterised in that:Include the following steps:
The symbolic feature of step 1, extraction texture image f (x, y);
The contrast difference energy feature of step 2, extraction texture image f (x, y);
The center pixel feature of step 3, extraction texture image f (x, y);
Step 4, symbolic feature, contrast difference energy feature and center pixel feature by the texture image f (x, y) of extraction
It is integrated, obtains the symbol energy center pixel histogram feature of texture image f (x, y), and establish histogram;
Step 5 classifies to histogram using chi-Square measure and nearest neighbor classifier.
2. a kind of image texture sorting technique based on contrast local binary patterns according to claim 1, feature
It is:The extraction of the symbolic feature of texture image f (x, y) extracts description using local binary patterns in the step 1.
3. a kind of image texture sorting technique based on contrast local binary patterns according to claim 1, feature
It is:The specific extracting method of the symbolic feature of texture image f (x, y) is in the step 1:
Step 1.1, the module for defining a 3*3 select pixel centered on a pixel, compare imago in this on that module
Element and 8 neighborhood territory pixel g of surroundingPBetween size, the use 0 less than center pixel indicates, is greater than or equal to the use 1 of center pixel
It indicates, then symbolic feature can be expressed as:
Step 1.2, according to a fixed direction by symbolic feature be considered as end to end annular, obtain the mould of equal value of LBP
Formula, with 0 to 1 transition times U (GP) indicate the equivalent formulations, then:
U (the G of step 1.3, the equivalent formulations for obtaining step 1.2P) rotated to obtain the LBP for rotating the not equivalent formulations of property,
It can specifically be expressed as:
The symbolic feature histogram H of step 1.4, texture image f (x, y)iIt is represented by:
4. a kind of image texture sorting technique based on contrast local binary patterns according to claim 1, feature
It is:The specific method of the contrast difference energy feature of extraction texture image f (x, y) is in the step 2:
Step 2.1, the contrast difference energy feature CDEF for calculating each pixel:One is defined on texture image f (x, y)
The module of 3*3 selects pixel g centered on a pixel on that modulec, contrast difference energy feature CDEF can be by following formula
It is calculated:Wherein gPFor gcNeighborhood territory pixel, 1≤P≤9;
Step 2.2, setting difference value FD:Wherein max is the maximum value of image pixel, and min is image slices
The minimum value of element;
Step 2.3, the section CV that bin histogram is set:13 critical value CV are set0,CV1..., CV12, value is by following formula meter
It calculates:
CVi=H (x)+min i ∈ [0,12]
5. a kind of image texture sorting technique based on contrast local binary patterns according to claim 1, feature
It is:The specific method of the center pixel feature F (x, y) of extraction texture image f (x, y) is in the step 3:F (x, y)=f
(x, y) modn, wherein f (x, y) indicate the texture image, n > 2.
6. a kind of image texture sorting technique based on contrast local binary patterns according to claim 5, feature
It is:The value of the n is 12.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810379373.2A CN108564095A (en) | 2018-04-25 | 2018-04-25 | A kind of image texture sorting technique based on contrast local binary patterns |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810379373.2A CN108564095A (en) | 2018-04-25 | 2018-04-25 | A kind of image texture sorting technique based on contrast local binary patterns |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108564095A true CN108564095A (en) | 2018-09-21 |
Family
ID=63536516
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810379373.2A Pending CN108564095A (en) | 2018-04-25 | 2018-04-25 | A kind of image texture sorting technique based on contrast local binary patterns |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108564095A (en) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160110590A1 (en) * | 2014-10-15 | 2016-04-21 | University Of Seoul Industry Cooperation Foundation | Facial identification method, facial identification apparatus and computer program for executing the method |
CN106408029A (en) * | 2016-09-28 | 2017-02-15 | 河南科技大学 | Image texture classification method based on structural difference histogram |
CN107832714A (en) * | 2017-11-14 | 2018-03-23 | 腾讯科技(上海)有限公司 | Live body discrimination method, device and storage device |
-
2018
- 2018-04-25 CN CN201810379373.2A patent/CN108564095A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160110590A1 (en) * | 2014-10-15 | 2016-04-21 | University Of Seoul Industry Cooperation Foundation | Facial identification method, facial identification apparatus and computer program for executing the method |
CN106408029A (en) * | 2016-09-28 | 2017-02-15 | 河南科技大学 | Image texture classification method based on structural difference histogram |
CN107832714A (en) * | 2017-11-14 | 2018-03-23 | 腾讯科技(上海)有限公司 | Live body discrimination method, device and storage device |
Non-Patent Citations (1)
Title |
---|
JIE SIMA ETC: "Extended Contrast Local Binary Pattern for Texture Classification", 《INTERNATIONAL JOURNAL OF NEW TECHNOLOGY AND RESEARCH (IJNTR)》 * |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106462746B (en) | Analyzing digital holographic microscopy data for hematology applications | |
Yuan et al. | Bag-of-words and object-based classification for cloud extraction from satellite imagery | |
CN105488809B (en) | Indoor scene semantic segmentation method based on RGBD descriptors | |
CN108549870A (en) | A kind of method and device that article display is differentiated | |
Ghosh et al. | Unsupervised grow-cut: cellular automata-based medical image segmentation | |
CN105513066B (en) | It is a kind of that the generic object detection method merged with super-pixel is chosen based on seed point | |
Deng et al. | Cloud detection in satellite images based on natural scene statistics and gabor features | |
CN107909102A (en) | A kind of sorting technique of histopathology image | |
CN108073940B (en) | Method for detecting 3D target example object in unstructured environment | |
Wu et al. | Natural scene text detection by multi-scale adaptive color clustering and non-text filtering | |
CN112396619A (en) | Small particle segmentation method based on semantic segmentation and internally complex composition | |
Naqvi et al. | Feature quality-based dynamic feature selection for improving salient object detection | |
Wang et al. | Superpixel-level target discrimination for high-resolution SAR images in complex scenes | |
CN108288265A (en) | A kind of segmentation and sorting technique towards HCC pathological image nucleus | |
CN109271997B (en) | Image texture classification method based on skip subdivision local mode | |
CN106548195A (en) | A kind of object detection method based on modified model HOG ULBP feature operators | |
CN109741351A (en) | A kind of classification responsive type edge detection method based on deep learning | |
Akbar et al. | Tumor localization in tissue microarrays using rotation invariant superpixel pyramids | |
CN105844299B (en) | A kind of image classification method based on bag of words | |
CN110069648A (en) | A kind of image search method and device | |
CN111832463A (en) | Deep learning-based traffic sign detection method | |
Guo et al. | Is local dominant orientation necessary for the classification of rotation invariant texture? | |
Kavitha et al. | A robust script identification system for historical Indian document images | |
Backes | Upper and lower volumetric fractal descriptors for texture classification | |
CN105574880A (en) | Color image segmentation method based on exponential moment pixel classification |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180921 |
|
RJ01 | Rejection of invention patent application after publication |