CN106815845A - Color image segmentation method based on pixels probability density classification - Google Patents
Color image segmentation method based on pixels probability density classification Download PDFInfo
- Publication number
- CN106815845A CN106815845A CN201611210313.5A CN201611210313A CN106815845A CN 106815845 A CN106815845 A CN 106815845A CN 201611210313 A CN201611210313 A CN 201611210313A CN 106815845 A CN106815845 A CN 106815845A
- Authority
- CN
- China
- Prior art keywords
- pixel
- super
- image
- refer
- entropy
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10024—Color image
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a kind of color image segmentation method based on pixels probability density classification, split original color image with entropy rate super-pixel generation method first, obtain super-pixel image；Secondly, super-pixel energy feature and entropy feature based on pixels probability density gradient are calculated；Then, with two dimensionEntropy carries out just segmentation to the super-pixel image for generating；Finally, useGrader carries out image segmentation.Test result indicate that, the method for the present invention considers between color component correlation and utilization is high performance due to introducing super-pixel during the feature for calculating super-pixelGrader is split so that the precision of image segmentation is greatly improved with speed.
Description
Technical field
Image partition method the present invention relates to be based on Threshold segmentation, it is more particularly to a kind of to be based on pixels probability density classification
Color image segmentation method, belong to Digital Image Segmentation technical field.
Background technology
With the development of science and technology, people increasingly increase the demand of information, how effectively to be picked out from magnanimity information
Useful information is the key issue of information processing.Image procossing is also evolving as the most common form of information processing,
Image segmentation is as the key link of image procossing so as to receive extensive concern.Generally, for piece image people only
Interested in a certain specific part, this is specifically partly called object, and remainder is called background.In multimedia signal processing
In, image segmentation is often essential, the purpose is to distinguish object and background in image, to enter to subject area
Row research.Image segmentation is widely used in many fields, such as medical science, military affairs, industry.Although existing various images point
Segmentation method, but due to the complexity of image, still it is suitable for all different types of figures without a kind of dividing method of standard at present
Picture, therefore, image Segmentation Technology is still one of focus of current research.
In recent years, scholar is studied with reference to image segmentation with specific theory, it is proposed that such as rim detection, region life
It is long with split degree, Threshold segmentation, feature space cluster, neutral net, based on svm classifier, based on various figures such as statistical modeling
As dividing method, wherein, it is most widely used with threshold segmentation method.The characteristics of threshold segmentation method is based on image space, will
Image is converted into bianry image, calculates threshold value, if image is divided into stem portion by threshold value, object is extracted from background
Come.This method needs to calculate the histogram of image, each one region of histogrammic peak value correspondence, using threshold value by image
Pixel is divided into different classifications, and then by extracted region interested out.Additionally, being global threshold when threshold value is constant
Value method, is otherwise local threshold method, when background light is uneven, is difficult to be partitioned into target area using global threshold method
Domain, and this defect can then be made up using local threshold method.Threshold segmentation method is simple and effective, but the method is remained
Following deficiency：First, it is easily affected by noise, under very noisy interference, it is impossible to obtain gratifying effect；Second, time
Efficiency is very low during the pixel gone through in tonal range, and computationally intensive；3rd, do not account for image space feature, i.e., it cannot be guaranteed that
The segmentation of adjacent area.
The content of the invention
The present invention is to solve the above-mentioned technical problem existing for prior art, it is proposed that one kind considers color of image
The color image segmentation method based on pixels probability density classification of correlation between component.
Technical solution of the invention is：A kind of color image segmentation method based on pixels probability density classification, its
It is characterised by：Specifically include following steps：
Agreement：Refer to original color image；Refer to super-pixel image；Refer toIn super-pixel total number；Refer toProbability
Density gradient；Refer to entropy rate super-pixel generation method；Finger is based on'sIn energy；Finger is based on'sIn
Entropy；Refer toGray value；Refer toAverage gray value；Refer to Image Classifier；
A. initial setting up
Obtain original color image and Initialize installation；
B. original color image super-pixel segmentation
WithMethod is to original color imageSegmentation, generates super-pixel image；
C. super-pixel feature calculation
C.1 calculated according to following formulaIn（Property field is with the pixel in the space of spatial domain, wherein face
ColorTake）：
Wherein,,,,Refer to
Unitization constant,Refer to the bandwidth in image spatial domain and feature domain space,WithRespectively plane of delineation bandwidth,
Brightness bandwidth and color bandwidth；
C.2 by the probability density gradient of original color imageIt is assigned to the super-pixel image for having generated, calculate excess of export picture
The energy feature of elementAnd entropy feature；
D. super-pixel two-dimensional imageSplit at the beginning of entropy
D.1 calculated according to following formulaIn average gray level：
；
Wherein,Represent theGray value in individual super-pixel region；Represent theIndividual super-pixel area
In domainThe abscissa of individual super-pixel point；Represent theIn individual super-pixel regionThe ordinate of individual super-pixel point；
D.2 setWithThe gray level of composition to for, noteThe number of appearance is, thenTwo dimension joint
Probability density is：
；
D.3 according to following formula, two dimension is calculated respectivelyThe target and background of entropy：
Wherein,It is the gray level of super-pixel image；For or not 1 positive number；It is threshold vector；It is destination probability；It is background probability；
D.4 discriminant function is defined as follows：
WhenWhen taking maximum, you can obtain optimal threshold：
D.5 just segmentation result is obtained using optimal threshold, is chosenIndividual object pixel andIndividual background pixel is used as training sample
This, all training samples form complete training set, and residual image pixel forms test set；
e. Model training
Trained using the training data chosenModel；
f. Category of model
The class label of test set is predicted, using two dimensionEntropy threshold obtain training set class label, merge test set and
The class label of training set forms class label vector, used as the segmentation result of image.
The present invention splits original color image with entropy rate super-pixel generation method first, obtains super-pixel image；Secondly, meter
Calculate super-pixel energy feature and entropy feature based on pixels probability density gradient；Then, with two dimensionEntropy opposite
Into super-pixel image carry out just segmentation；Finally, useGrader carries out image segmentation.Test result indicate that, it is of the invention
Method considers between color component correlation and utilization is high performance due to introducing super-pixel during the feature for calculating super-pixelGrader is split so that the precision of image segmentation is greatly improved with speed.
Compared with prior art, the invention has the advantages that：
First, the application of super-pixel reduces the complexity of feature extraction, reduces time-consuming；
Second, during construction local feature, it is contemplated that correlation union textural characteristics, color characteristic, shape facility between color component
In one, super-pixel feature is preferably featured；
3rd, high-performanceApplication so that the segmentation precision and speed of image are substantially better than other traditional classifiers.
Brief description of the drawings
Fig. 1 is that embodiment of the present invention super-pixel image generates result figure.
Fig. 2 is the probability density Gradient Features result figure of embodiment of the present invention original color image.
Fig. 3 is the energy feature result figure of embodiment of the present invention super-pixel image.
Fig. 4 is the entropy characteristic results figure of embodiment of the present invention super-pixel image.
Fig. 5 is embodiment of the present invention super-pixel two-dimensional imageThe first segmentation result figure of entropy.
Fig. 6 is the embodiment of the present inventionSegmentation and comparative result figure.
Fig. 7 is the flow chart of the embodiment of the present invention.
Specific embodiment
The method of the present invention includes four-stage altogether：Original color image super-pixel segmentation, super-pixel feature calculation, super picture
Plain two-dimensional imageSplit at the beginning of entropy, utilizeModel carries out pixel classifications.
Agreement：Refer to original color image；Refer to super-pixel image；Refer toIn super-pixel total number；Refer toIt is general
Rate density gradient；Refer to entropy rate super-pixel generation method；Finger is based on'sIn energy；Finger is based on'sIn
Entropy；Refer toGray value；Refer toAverage gray value；Refer to Image Classifier；
A. initial setting up
Obtain original color image and Initialize installation；
B. original color image super-pixel segmentation
WithMethod is to original color imageSegmentation, generates super-pixel image；
C. super-pixel feature calculation
C.1 calculated according to following formulaIn（Property field is with the pixel in the space of spatial domain, wherein face
ColorTake）：
Wherein,,,,
Refer to unitization constant,Refer to the bandwidth in image spatial domain and feature domain space,WithRespectively plane of delineation band
Wide, brightness bandwidth and color bandwidth；
C.2 by the probability density gradient of original color imageIt is assigned to the super-pixel image for having generated, calculate excess of export picture
The energy feature of elementAnd entropy feature；
D. super-pixel two-dimensional imageSplit at the beginning of entropy
D.1 calculated according to following formulaIn average gray level：
；
Wherein,Represent theGray value in individual super-pixel region；Represent theIndividual super-pixel area
In domainThe abscissa of individual super-pixel point；Represent theIn individual super-pixel regionThe ordinate of individual super-pixel point；
D.2 setWithThe gray level of composition to for, noteThe number of appearance is, thenTwo dimension joint
Probability density is：
；
D.3 according to following formula, two dimension is calculated respectivelyThe target and background of entropy：
Wherein,It is the gray level of super-pixel image；For or not 1 positive number；It is threshold vector；It is destination probability；It is background probability；
D.4 discriminant function is defined as follows：
WhenWhen taking maximum, you can obtain optimal threshold：
D.5 just segmentation result is obtained using optimal threshold, is chosenIndividual object pixel andIndividual background pixel is used as training sample
This, all training samples form complete training set, and residual image pixel forms test set；
e. Model training
Trained using the training data chosenModel；
f. Category of model
The class label of test set is predicted, using two dimensionEntropy threshold obtain training set class label, merge test set and
The class label of training set forms class label vector, used as the segmentation result of image.
Experiment test and parameter setting：
Experiment is performed under MATLAB 7.12.0 (R2011a) environment, and what experiment was related to is that resolution ratio is 255*170 pictures
Element, 300*225 pixels, the coloured image of 300*420 pixels, involved image comes from three databases, respectively
The image data base in Berkeley partition datas storehouse (BSD), segmentation evaluation database (SED) and the research object identification of Cambridge Microsoft
(MSRC)。
Embodiment of the present invention generation super-pixel image result is as shown in Figure 1.
The probability density Gradient Features result that the embodiment of the present invention calculates original color image is as shown in Figure 2.
The energy feature result of embodiment of the present invention super-pixel image is as shown in Figure 3.
The entropy characteristic results of embodiment of the present invention super-pixel image are as shown in Figure 4.
Embodiment of the present invention super-pixel two-dimensional imageJust segmentation result figure is as shown in Figure 5 for entropy.
The embodiment of the present inventionSegmentation and comparative result figure are as shown in Figure 6.
Claims (1)
1. a kind of color image segmentation method based on pixels probability density classification, it is characterised in that follow the steps below：
Agreement：Refer to original color image；Refer to super-pixel image；Refer toIn super-pixel total number；Refer toProbability it is close
Degree gradient；Refer to entropy rate super-pixel generation method；Finger is based on'sIn energy；Finger is based on'sIn entropy
Value；Refer toGray value；Refer toAverage gray value；Refer to Image Classifier；
A. initial setting up
Obtain original color image and Initialize installation；
B. original color image super-pixel segmentation
WithMethod is to original color imageSegmentation, generates super-pixel image；
C. super-pixel feature calculation
C.1 calculated according to following formulaIn,Property field is with the pixel in the space of spatial domain, wherein face
ColorTake：
Wherein,,,,Refer to single
Positionization constant,Refer to the bandwidth in image spatial domain and feature domain space,WithRespectively plane of delineation bandwidth, bright
Degree bandwidth and color bandwidth；
C.2 by the probability density gradient of original color imageIt is assigned to the super-pixel image for having generated, calculate excess of export picture
The energy feature of elementAnd entropy feature；
D. super-pixel two-dimensional imageSplit at the beginning of entropy
D.1 calculated according to following formulaIn average gray level：
；
Wherein,Represent theGray value in individual super-pixel region；Represent theIndividual super-pixel area
In domainThe abscissa of individual super-pixel point；Represent theIn individual super-pixel regionThe ordinate of individual super-pixel point；
D.2 setWithThe gray level of composition to for, noteThe number of appearance is, thenTwo dimension joint
Probability density is：
；
D.3 according to following formula, two dimension is calculated respectivelyThe target and background of entropy：
Wherein,It is the gray level of super-pixel image；For or not 1 positive number；It is threshold vector；It is destination probability；It is background probability；
D.4 discriminant function is defined as follows：
WhenWhen taking maximum, you can obtain optimal threshold：
D.5 just segmentation result is obtained using optimal threshold, is chosenIndividual object pixel andIndividual background pixel is used as training sample
This, all training samples form complete training set, and residual image pixel forms test set；
e. Model training
Trained using the training data chosenModel；
f. Category of model
The class label of test set is predicted, using two dimensionEntropy threshold obtain training set class label, merge test set and
The class label of training set forms class label vector, used as the segmentation result of image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611210313.5A CN106815845A (en) | 2016-12-24 | 2016-12-24 | Color image segmentation method based on pixels probability density classification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611210313.5A CN106815845A (en) | 2016-12-24 | 2016-12-24 | Color image segmentation method based on pixels probability density classification |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106815845A true CN106815845A (en) | 2017-06-09 |
Family
ID=59110086
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611210313.5A Pending CN106815845A (en) | 2016-12-24 | 2016-12-24 | Color image segmentation method based on pixels probability density classification |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106815845A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107481261A (en) * | 2017-07-31 | 2017-12-15 | 中国科学院长春光学精密机械与物理研究所 | A kind of color video based on the tracking of depth prospect scratches drawing method |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120275702A1 (en) * | 2011-04-29 | 2012-11-01 | Cuneyt Oncel Tuzel | Method for Segmenting Images Using Superpixels and Entropy Rate Clustering |
| CN105574880A (en) * | 2015-12-28 | 2016-05-11 | 辽宁师范大学 | Color image segmentation method based on exponential moment pixel classification |
-
2016
- 2016-12-24 CN CN201611210313.5A patent/CN106815845A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120275702A1 (en) * | 2011-04-29 | 2012-11-01 | Cuneyt Oncel Tuzel | Method for Segmenting Images Using Superpixels and Entropy Rate Clustering |
| CN105574880A (en) * | 2015-12-28 | 2016-05-11 | 辽宁师范大学 | Color image segmentation method based on exponential moment pixel classification |
Non-Patent Citations (2)
| Title |
|---|
| 王钦琰: ""基于统计建模的彩色图像分割算法研究"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 纪超等: ""基于超像素的物体似然概率计算模型研究"", 《红外与激光工程》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107481261A (en) * | 2017-07-31 | 2017-12-15 | 中国科学院长春光学精密机械与物理研究所 | A kind of color video based on the tracking of depth prospect scratches drawing method |
| CN107481261B (en) * | 2017-07-31 | 2020-06-16 | 中国科学院长春光学精密机械与物理研究所 | Color video matting method based on depth foreground tracking |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Genovese et al. | ASIC and FPGA implementation of the Gaussian mixture model algorithm for real-time segmentation of high definition video | |
| CN104599275B (en) | The RGB-D scene understanding methods of imparametrization based on probability graph model | |
| CN107564022B (en) | Saliency detection method based on Bayesian Fusion | |
| CN108121991B (en) | Deep learning ship target detection method based on edge candidate region extraction | |
| WO2018023734A1 (en) | Significance testing method for 3d image | |
| CN104850850B (en) | A kind of binocular stereo vision image characteristic extracting method of combination shape and color | |
| CN105761238B (en) | A method of passing through gray-scale statistical data depth information extraction well-marked target | |
| CN106991686B (en) | A kind of level set contour tracing method based on super-pixel optical flow field | |
| CN107038416B (en) | Pedestrian detection method based on binary image improved HOG characteristics | |
| CN107169417B (en) | RGBD image collaborative saliency detection method based on multi-core enhancement and saliency fusion | |
| CN104657980A (en) | Improved multi-channel image partitioning algorithm based on Meanshift | |
| CN111127360B (en) | Gray image transfer learning method based on automatic encoder | |
| CN107862680B (en) | Target tracking optimization method based on correlation filter | |
| CN105513080B (en) | A kind of infrared image target Salience estimation | |
| CN110782442B (en) | Image artificial fuzzy detection method based on multi-domain coupling | |
| JP2004062605A (en) | Scene identification method and device, and program | |
| CN106203448B (en) | A kind of scene classification method based on Nonlinear Scale Space Theory | |
| CN107067037A (en) | A kind of method that use LLC criterions position display foreground | |
| Yuan et al. | Superpixels with content-adaptive criteria | |
| CN104504715A (en) | Image segmentation method based on local quaternion-moment characteristic | |
| CN109241932A (en) | A kind of thermal infrared human motion recognition method based on movement variogram phase property | |
| CN106815845A (en) | Color image segmentation method based on pixels probability density classification | |
| CN105574880A (en) | Color image segmentation method based on exponential moment pixel classification | |
| CN115223033A (en) | Synthetic aperture sonar image target classification method and system | |
| CN107085725A (en) | A kind of method that image-region is clustered by the LLC based on adaptive codebook |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20170609 |