CN103390276B - A kind of color image processing method based on three-dimensional Gaussian Cloud transform and device - Google Patents
A kind of color image processing method based on three-dimensional Gaussian Cloud transform and device Download PDFInfo
- Publication number
- CN103390276B CN103390276B CN201310309949.5A CN201310309949A CN103390276B CN 103390276 B CN103390276 B CN 103390276B CN 201310309949 A CN201310309949 A CN 201310309949A CN 103390276 B CN103390276 B CN 103390276B
- Authority
- CN
- China
- Prior art keywords
- dimensional gaussian
- cloud
- dimensional
- degree
- component
- 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.)
- Expired - Fee Related
Links
- 238000003672 processing method Methods 0.000 title claims abstract description 8
- 239000003086 colorant Substances 0.000 claims abstract description 38
- 238000000034 method Methods 0.000 claims abstract description 25
- 230000015572 biosynthetic process Effects 0.000 claims abstract description 22
- 238000003786 synthesis reaction Methods 0.000 claims abstract description 22
- 238000000605 extraction Methods 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 2
- 241000208340 Araliaceae Species 0.000 claims 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 claims 1
- 235000003140 Panax quinquefolius Nutrition 0.000 claims 1
- 235000008434 ginseng Nutrition 0.000 claims 1
- 235000019580 granularity Nutrition 0.000 description 7
- 230000011218 segmentation Effects 0.000 description 6
- 239000000284 extract Substances 0.000 description 4
- 238000003709 image segmentation Methods 0.000 description 4
- 239000002245 particle Substances 0.000 description 4
- 230000009466 transformation Effects 0.000 description 4
- 238000006243 chemical reaction Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000001149 cognitive effect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 239000008187 granular material Substances 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000739 chaotic effect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 239000000428 dust Substances 0.000 description 1
- 238000003708 edge detection Methods 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 239000013535 sea water Substances 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of color image processing method based on three-dimensional Gaussian Cloud transform and device, described method includes: by extracting the three primary colours color value of each pixel of coloured image, form the color value data sample set of described coloured image；According to default initial three-dimensional Gaussian cloud quantity m, described color value data sample set is synthesized m three-dimensional Gaussian cloud；Successively the indistinct degree of m three-dimensional Gaussian cloud is compared with ambiguity degree threshold value, to determine whether there is its ambiguity degree three-dimensional Gaussian cloud more than ambiguity degree threshold value；If there is its ambiguity degree three-dimensional Gaussian cloud more than ambiguity degree threshold value, then according to the mode that described quantity m is gradually subtracted, repeat the synthesis process of three-dimensional Gaussian cloud of color value data sample set and three-dimensional Gaussian cloud ambiguity degree and ambiguity degree threshold ratio compared with process, until the indistinct degree of all three-dimensional Gaussian clouds is respectively less than is equal to described ambiguity degree threshold value；Export its ambiguity degree and be respectively less than all of three-dimensional Gaussian cloud equal to described ambiguity degree threshold value.
Description
Technical field
The present invention relates to image processing field, particularly to a kind of method utilizing three-dimensional Gaussian Cloud transform that coloured image is processed and relevant apparatus thereof.
Background technology
Granule Computing is that things is indicated with simulating human from different grain size, different levels, analyzes and the method for reasoning by research, is an important directions of intelligent information processing technology research in artificial intelligence.In human cognitive, the expression of many granularities concept, generation and suitable particle size selects the difficult problem that always Granule Computing faces.Cloud model is a cognitive model studying qualitative, quantitative conversion based on probability theory.Cloud model is with three kinds of numerical characteristics: expectation, entropy and super entropy, characterize the intension of a qualitativing concept, it is contemplated to be the data sample that can represent concept, entropy is the tolerance of conceptual type, the scope of data that reflection concept is contained, super entropy is the uncertainty measure of conceptual type, the uncertainty on reflection scope of data border.
Gauss distribution is most important distribution in theory of probability, rely on data statistics and the concept that takes out usually has the feature of " broad in the middle, two is little ", standard deviation in Gauss distribution can be used to characterize the granularity of concept, cloud model introduces super entropy and weighs the uncertainty of concept granularity, the i.e. uncertainty of standard deviation, therefore Gauss cloud model is the most frequently used a kind of cloud model.
Cloud model realizes a conversion between concept connotation and extension by forward cloud algorithm and reverse cloud algorithm.One group of data sample can be converted to three numerical characteristics of a basic conception by reverse cloud algorithm, but the premise of this algorithm acquiescence is, given all data samples correspond to same concept extension in same granularity and characterize, and can not solve the generation problem of many granularities, many concepts in whole Problem Areas.
In color images, coloured image can be divided into several target area according to the color value of pixel.At present; dividing method based on Gaussian transformation is mainly according to the corresponding probit size in each Gauss distribution of each pixel in image; determine which target area pixel belongs to; thus when image is split; intersection in two regions usually there will be factitious sawtooth, it is impossible to embodies the range of indeterminacy between target area.
Summary of the invention
It is an object of the invention to provide a kind of color image processing method based on three-dimensional Gaussian Cloud transform and device, can preferably solve the problem of the overlapping confusion of transitional region during color images.
According to an aspect of the invention, it is provided a kind of color image processing method based on three-dimensional Gaussian Cloud transform, including:
A) by extracting the three primary colours color value of each pixel of coloured image, the color value data sample set of described coloured image is formed；
B) according to default initial three-dimensional Gaussian cloud quantity m, described color value data sample set is synthesized m three-dimensional Gaussian cloud；
C) successively the indistinct degree of m three-dimensional Gaussian cloud is compared with ambiguity degree threshold value, to determine whether there is its ambiguity degree three-dimensional Gaussian cloud more than ambiguity degree threshold value；
D) if there is its ambiguity degree three-dimensional Gaussian cloud more than ambiguity degree threshold value, then according to the mode that described quantity m is gradually subtracted, repeat the synthesis process of three-dimensional Gaussian cloud of color value data sample set and three-dimensional Gaussian cloud ambiguity degree and ambiguity degree threshold ratio compared with process, until the indistinct degree of all three-dimensional Gaussian clouds is respectively less than is equal to described ambiguity degree threshold value；
E) export its ambiguity degree and be respectively less than all of three-dimensional Gaussian cloud equal to described ambiguity degree threshold value.
Preferably, described step B) including:
B1) according to quantity m of default initial three-dimensional Gaussian cloud, the color value data sample of described coloured image is converted into m three-dimensional Gaussian component；
B2) utilize described m three-dimensional Gaussian component, calculate the three-dimensional Gaussian YUNSHEN number that each three-dimensional Gaussian component is corresponding, thus obtain m three-dimensional Gaussian cloud.
Preferably, described step B1) including:
B11) frequency distribution of three primary colours in described color value data sample set is added up；
B12) according to described quantity m, set respectively and include that the initial expectation of three primary colours, primary standard be poor, the m group initial parameter of initial weight, and utilize described m group initial parameter, determine the m group estimation parameter estimating expectation, estimated standard deviation, estimation weights including three primary colours；
B13) utilize the frequency distribution of described three primary colours and described m group initial parameter, calculate the initial function value of goal-selling function, and utilize the frequency distribution of described three primary colours and described m group to estimate parameter, calculate the estimation function value of goal-selling function；
B14) difference between described estimation function value and described initial function value is calculated, if described difference is more than or equal to predictive error threshold value, then use described estimation parameter to replace described initial parameter, and repeat step B13), until described difference is less than predictive error threshold value；
B15) export described difference and estimate parameter less than the m group of predictive error threshold value, and described m group is estimated the parameter m group distributed constant as m three-dimensional Gaussian component, wherein, that estimates the three primary colours in parameter estimates that expectation, estimated standard deviation, estimation weights are respectively as the distribution expectation of the three primary colours in distributed constant, standard deviation, distribution weights.
Preferably, described step B2) including:
B21) pantograph ratio of m standard deviation of m three-dimensional Gaussian component is calculated respectively；
B22) utilize pantograph ratio and the distributed constant of described m three-dimensional Gaussian component, determine the cloud comprising three primary colours expectation that each three-dimensional Gaussian component is corresponding, entropy, super entropy, the three-dimensional Gaussian YUNSHEN number of ambiguity degree, thus obtain m three-dimensional Gaussian cloud.
Preferably for the kth three-dimensional Gaussian component in described m three-dimensional Gaussian component, the calculation procedure of the pantograph ratio of its standard deviation includes:
In described m three-dimensional Gaussian component, the overlapping degree overlapping with other three-dimensional Gaussian component by calculating kth three-dimensional Gaussian component respectively, obtain the l the three-dimensional Gaussian component maximum with kth three-dimensional Gaussian component degree of overlapping；
Described kth three-dimensional Gaussian component and described the l three-dimensional Gaussian component are carried out the scaling of equal proportion, obtains the pantograph ratio making the described kth three-dimensional Gaussian component scaled and described the l three-dimensional Gaussian component not overlap.
Preferably, described ambiguity degree is N times of super entropy and entropy ratio, and N is positive integer.
Preferably, also include:
Utilize all of three-dimensional Gaussian cloud exported, described coloured image is split.
According to a further aspect in the invention, it is provided that a kind of color image processing apparatus based on three-dimensional Gaussian Cloud transform, including:
Color value extraction module, for the three primary colours color value by extracting each pixel of coloured image, forms the color value data sample set of described coloured image；
Three-dimensional Gaussian cloud synthesis module, for according to default initial three-dimensional Gaussian cloud quantity m, described color value data sample set is synthesized m three-dimensional Gaussian cloud, and successively the indistinct degree of m three-dimensional Gaussian cloud is compared with ambiguity degree threshold value, to determine whether there is its ambiguity degree three-dimensional Gaussian cloud more than ambiguity degree threshold value, if there is its ambiguity degree three-dimensional Gaussian cloud more than ambiguity degree threshold value, then according to the mode that described quantity m is gradually subtracted, repeat the synthesis process of three-dimensional Gaussian cloud of color value data sample set and three-dimensional Gaussian cloud ambiguity degree and ambiguity degree threshold ratio compared with process, until the indistinct degree of all three-dimensional Gaussian clouds is respectively less than equal to described ambiguity degree threshold value；
Three-dimensional Gaussian cloud output module, is respectively less than all of three-dimensional Gaussian cloud equal to described ambiguity degree threshold value for exporting its ambiguity degree.
Preferably, described three-dimensional Gaussian cloud synthesis module includes:
Three-dimensional Gaussian component synthon module, for quantity m according to default initial three-dimensional Gaussian cloud, is converted into m three-dimensional Gaussian component by the color value data sample of described coloured image；
Three-dimensional Gaussian cloud computing submodule, is used for utilizing described m three-dimensional Gaussian component, calculates the three-dimensional Gaussian YUNSHEN number that each three-dimensional Gaussian component is corresponding, thus obtain m three-dimensional Gaussian cloud.
Preferably, also include:
Color images module, for utilizing all of three-dimensional Gaussian cloud exported, splits described coloured image.
Compared with prior art, the beneficial effects of the present invention is:
The present invention extracts by the color value data sample set of image carries out Gaussian cloud transformation and Gauss cloud, synthesize suitable multiple Gauss cloud, and the multiple Gauss clouds obtained by utilizing carry out color images effectively, in terms of the transitional region, uncertain rim detection of image, obtain more preferable segmentation effect.
Accompanying drawing explanation
Fig. 1 is a kind of based on three-dimensional Gaussian Cloud transform the color image processing method flow chart that the embodiment of the present invention provides；
Fig. 2 a is the width artwork that the embodiment of the present invention provides；
Fig. 2 b is three the three-dimensional Gaussian clouds formed by Gaussian cloud transformation；
Fig. 2 c is the result schematic diagram utilizing described three three-dimensional Gaussian clouds to split coloured image；
Fig. 2 d is the segmentation object that the minimum Gauss cloud of ambiguity degree is corresponding；
Fig. 2 e is the uncertain edge schematic diagram in image；
Fig. 3 is a kind of based on three-dimensional Gaussian Cloud transform the color image processing apparatus structured flowchart that the embodiment of the present invention provides.
Detailed description of the invention
Below in conjunction with accompanying drawing to a preferred embodiment of the present invention will be described in detail, it will be appreciated that preferred embodiment described below is merely to illustrate and explains the present invention, is not intended to limit the present invention.
nullThe present invention first passes through the three primary colours color value extracting each pixel of coloured image，Form the color value data sample set of described coloured image，Then gauss hybrid models (Gaussian Mixture Model is utilized，And expectation maximization algorithm (Expectation Maximization GMM)，EM)，By described color value data sample set according to default quantity m,M Gaussian component in synthesis gauss hybrid models GMM，And m three-dimensional Gaussian component conversion in gauss hybrid models GMM is had m three-dimensional Gaussian cloud of different meaning，Finally，According to ambiguity degree，The quantity of optimization gauss cloud，And carry out image segmentation，Specifically，Successively the indistinct degree of m three-dimensional Gaussian cloud is compared with ambiguity degree threshold value，To determine whether there is its ambiguity degree three-dimensional Gaussian cloud more than ambiguity degree threshold value，If there is its ambiguity degree three-dimensional Gaussian cloud more than ambiguity degree threshold value，Then according to the mode that described quantity m is gradually subtracted，Repeat the synthesis process of three-dimensional Gaussian cloud of color value data sample set and three-dimensional Gaussian cloud ambiguity degree and ambiguity degree threshold ratio compared with process，Until the indistinct degree of all three-dimensional Gaussian clouds is respectively less than equal to described ambiguity degree threshold value，Export its ambiguity degree and be respectively less than all of three-dimensional Gaussian cloud equal to described ambiguity degree threshold value.Relative to prior art, the present invention has more preferable segmentation effect at transitional region, the uncertain side edge detection mask of image.
The present invention includes from from key technologies such as three-dimensional Gaussian component to three-dimensional Gaussian cloud, the synthesis of three-dimensional Gaussian cloud and image segmentations.Wherein:
1, from three-dimensional Gaussian component to three-dimensional Gaussian cloud
The Gaussian component that any two utilizing gauss hybrid models and expectation maximization algorithm to obtain is intersected, using they current standard deviations as the maximum particle size parameter of three-dimensional Gaussian cloud, according to " associating weak between class, association is tight in class " principle, the expectation keeping them is constant, carry out equal proportion reduction, until their weak outer peripheral areas element is non-intersect, now can obtain the minimum particle size parameter of each three-dimensional Gaussian cloud, utilize three-dimensional Gaussian cloud particle degree excursion can obtain the entropy En and super entropy He of each three-dimensional Gaussian cloud, owing to carrying out equal proportion reduction, so their He/En is identical, it is referred to as ambiguity degree, its computing formula is as follows:
CD=N×He/En
Wherein, N is positive integer, it is preferred that N value is 3.
Described ambiguity degree can be used to weigh the dispersion degree of three-dimensional Gaussian cloud extension, is also the degree of three-dimensional Gaussian cloud distribution deviation three-dimensional Gaussian distribution.Ambiguity degree is 0, then three-dimensional Gaussian cloud extension converges, and builds consensus, is a ripe three-dimensional Gaussian cloud；Ambiguity degree is 1, then three-dimensional Gaussian cloud extension dissipates, it is difficult to build consensus, and is i.e. atomized.
2, the synthesis of three-dimensional Gaussian cloud and image segmentation
Three-dimensional Gaussian Cloud transform uses forward Gauss cloud algorithm to generate each pixel degree of certainty to three-dimensional Gaussian cloud, by the compared pixels point degree of certainty size to different three-dimensional Gaussian clouds, judge which three-dimensional Gaussian cloud it belongs to, this degree of certainty obtained that calculates has uncertainty, therefore can characterize the range of indeterminacy between adjacent three-dimensional Gauss cloud.
Below by way of Fig. 1 to Fig. 3, the present invention is carried out deep explanation.
Fig. 1 is a kind of based on three-dimensional Gaussian Cloud transform the color image processing method flow chart that the embodiment of the present invention provides, as it is shown in figure 1, step includes:
Step 101, by extract each pixel of coloured image three primary colours color value < red Ri, green Gi, blue Bi>, form color value data the sample set {<R of described coloured imagei,Gi,Bi> | i=1,2 ..., N}, wherein, Ri、Gi、BiValue is all between 0 to 255.
Step 102, according to default initial three-dimensional Gaussian cloud quantity m, described color value data sample set is synthesized m three-dimensional Gaussian cloud, wherein, m is an integer being more than 1 set in advance.
Described step 102 includes:
Step 102-1: by gauss hybrid models and expectation maximization algorithm, described color value data sample set is converted into m three-dimensional Gaussian component G (< μ x according to quantity m of default initial three-dimensional Gaussian cloudk,μyk,μzk>,σk),k=1,…,m。
Specifically, the frequency distribution of three primary colours in described color value data sample set is added up；According to described quantity m, set respectively and include that the initial expectation of three primary colours, primary standard be poor, the m group initial parameter of initial weight, and utilize described m group initial parameter, determine the m group estimation parameter including the estimation expectation of three primary colours, estimated standard deviation, estimation weights；Utilize the frequency distribution of described three primary colours and described m group initial parameter, calculate the initial function value of goal-selling function, and utilize the frequency distribution of described three primary colours and described m group to estimate parameter, calculate the estimation function value of goal-selling function；Calculate the difference between described estimation function value and described initial function value, if described difference is more than or equal to predictive error threshold value, described estimation parameter is then used to replace described initial parameter, and repeat step above-mentioned steps, until described difference is less than predictive error threshold value, export described difference and estimate parameter less than the m group of predictive error threshold value, and described m group is estimated the parameter m group distributed constant as m three-dimensional Gaussian component, wherein, estimate that the estimation of the three primary colours in parameter is expected, estimated standard deviation, estimate weights distribution expectation < the μ x respectively as the three primary colours in distributed constantk,μyk,μzk>, standard deviation sigmak, distribution weights.
Step 102-2: utilize described m three-dimensional Gaussian component G (< μ xk,μyk,μzk>,σk), k=1 ..., m, calculate the three-dimensional Gaussian YUNSHEN number that each three-dimensional Gaussian component is corresponding, thus obtain m three-dimensional Gaussian cloud C (< R_Exk,G_Exk,B_Exk>,Enk,Hek),k=1,…,m。
Specifically, calculate the pantograph ratio of m standard deviation of m three-dimensional Gaussian component respectively, utilize pantograph ratio and the distributed constant of described m three-dimensional Gaussian component, determine the cloud the comprising three primary colours expectation < R_Ex that each three-dimensional Gaussian component is correspondingk,G_Exk,B_Exk>, entropy Enk, super entropy Hek, ambiguity degree CDkThree-dimensional Gaussian YUNSHEN number, thus obtain m three-dimensional Gaussian cloud, wherein, the indistinct degree CD of each three-dimensional Gaussian cloudkFor super entropy HekWith entropy EnkN times of ratio, N is positive integer, it is preferable that N value is 3.Furtherly, by calculating the overlapping degree between each three-dimensional Gaussian component and other three-dimensional Gaussian component respectively, each three-dimensional Gaussian component is converted to three-dimensional Gaussian cloud.
Furtherly, calculate the overlapping degree between each three-dimensional Gaussian amount and other three-dimensional Gaussian components respectively, formed and overlap matrix: Calculate and obtain the overlapping degree of maximum of each three-dimensional Gaussian component.For the kth three-dimensional Gaussian component in described m three-dimensional Gaussian component, the overlapping degree overlapping with other three-dimensional Gaussian component by calculating described kth three-dimensional Gaussian component respectively, obtain the three-dimensional Gaussian component maximum with kth three-dimensional Gaussian component degree of overlapping, assume that the maximum degree of overlapping of kth three-dimensional Gaussian component is overlapping between the l three-dimensional Gaussian component, i.e. Max_IDk=max(IDk1,...,IDkn)=IDkl, then kth three-dimensional Gaussian component and the l three-dimensional Gaussian component being carried out the scaling of equal proportion, until not overlapping between them, i.e. meeting equation below:
Now: the three-dimensional Gaussian cloud that kth three-dimensional Gaussian component is corresponding is:
(<R_Exk,G_Exk,B_Exk>,(1+α)×σk/2,(1-α)*σk/6)
The indistinct degree of the three-dimensional Gaussian cloud that kth three-dimensional Gaussian component is corresponding is:
CDk=(1-α)/(1+α)
If for kth three-dimensional Gaussian component, calculating pantograph ratio α of its standard deviationk, then the Gauss YUNSHEN number of kth three-dimensional Gaussian cloud is:
R_Exk=μxk
G_Exk=μyk
B_Exk=μzk
Enk=(1+αk)×σk/2
Hek=(1-αk)×σk/6
CDk=(1-αk)/(1+αk)
Each three-dimensional Gaussian component granularity in three-dimensional is set as identical value En by described step 102, then in space, performance is a ball to kth three-dimensional Gaussian component in form, and sphere centre coordinate is (R_Exk,G_Exk,B_Exk), the radius of a ball is 3* σk。
Step 103, successively the indistinct degree of m three-dimensional Gaussian cloud is compared with ambiguity degree threshold value, to determine whether there is its ambiguity degree three-dimensional Gaussian cloud more than ambiguity degree threshold value.
If there is its ambiguity degree three-dimensional Gaussian cloud more than ambiguity degree threshold value in step 104, then according to the mode that described quantity m is gradually subtracted, repeating said steps 102 and described step 103, until the indistinct degree of all three-dimensional Gaussian clouds is respectively less than equal to described ambiguity degree threshold value, then export its ambiguity degree and be respectively less than all of three-dimensional Gaussian cloud equal to described ambiguity degree threshold value, to utilize all of three-dimensional Gaussian cloud exported, described coloured image is split.
Preferably, by ambiguity degree order from large to small, m three-dimensional Gaussian cloud is ranked up, and successively the indistinct degree CD of each three-dimensional Gaussian cloud is judged, if CDk> β, k=1 ..., m, β are previously given indistinct degree threshold value, then quantity m subtracts 1, and repeated execution of steps 102 and described step 103, and otherwise, output ambiguity degree is less than or equal to all of Gauss cloud of β.
Fig. 2 shows that a width resolution is the photochrome of 1024 × 717.A pair lovers is sitting in the background frame being made up of pink colour and white, and wherein sea water and the light through cloud layer sketch the contours of the artistic picture that whitewash is alternate.Every bit in image is regarded as a water dust, and it includes three attributes of red, green, blue, utilizes three-dimensional Gaussian Cloud transform that it is carried out concept extraction and foreground target extracts.Fig. 2 a is the width artwork that the embodiment of the present invention provides, and by extracting the three primary colours color value of each pixel of image, forms the color value data sample set of described image, then performs following steps:
1) initial parameter m is set；
2) utilize gauss hybrid models and expectation maximization algorithm to carry out parameter and generate m three-dimensional Gaussian component；
3) pantograph ratio α of m standard deviation of m Gaussian component is calculated respectivelyk, k=1,2 ..., m；
4) pantograph ratio α of described m standard deviation is utilizedkWith the distributed constant of described m three-dimensional Gaussian component, calculate m group three-dimensional Gaussian YUNSHEN number, three primary colours expectation < R_Ex respectivelyk,G_Exk,B_Exk>, entropy Enk, super entropy Hek, ambiguity degree CDkGauss YUNSHEN number, thus obtain m three-dimensional Gaussian cloud；
5) successively the indistinct degree of m three-dimensional Gaussian cloud is compared with ambiguity degree threshold value, to determine whether there is its ambiguity degree three-dimensional Gaussian cloud more than ambiguity degree threshold value；
6) if there is its ambiguity degree three-dimensional Gaussian cloud more than ambiguity degree threshold value, then according to the mode that described quantity m is gradually subtracted, step 2 is repeated) to step 5)；
7) exporting its ambiguity degree and be respectively less than all of three-dimensional Gaussian cloud equal to described ambiguity degree threshold value, in the present embodiment, it is 3 that ambiguity degree is respectively less than the number of the three-dimensional Gaussian cloud equal to described ambiguity degree threshold value, as shown in Figure 2 b.
8) utilize 3 the three-dimensional Gaussian clouds generated that described coloured image is split, obtain segmentation result as shown in Figure 2 c, overlapping owing to existing between three-dimensional Gaussian cloud, the uncertainty on this division border can utilize the super entropy of three-dimensional Gaussian cloud to calculate and obtain, therefore image based on Gaussian cloud transformation segmentation can divide the range of indeterminacy between concept, Fig. 2 d shows division black objects the most clearly, Fig. 2 e shows the uncertain edge in segmentation between target, and background is set to white.
Fig. 3 is a kind of based on three-dimensional Gaussian Cloud transform the color image processing apparatus structured flowchart that the embodiment of the present invention provides, as shown in Figure 3, including color value extraction module, three-dimensional Gaussian cloud synthesis module, three-dimensional Gaussian cloud output module and color images module, wherein:
Described color value extraction module, for the three primary colours color value by extracting each pixel of coloured image, forms color value data the sample set { < R of described coloured imagei,Gi,Bi> | i=1,2 ..., N}, wherein, Ri、Gi、BiValue is all between 0 to 255.
Described three-dimensional Gaussian cloud synthesis module is for according to default initial three-dimensional Gaussian cloud quantity m, described color value data sample set is synthesized m three-dimensional Gaussian cloud, and successively the indistinct degree of m three-dimensional Gaussian cloud is compared with ambiguity degree threshold value, to determine whether there is its ambiguity degree three-dimensional Gaussian cloud more than ambiguity degree threshold value, if there is its ambiguity degree three-dimensional Gaussian cloud more than ambiguity degree threshold value, then according to the mode that described quantity m is gradually subtracted, repeat the synthesis process of three-dimensional Gaussian cloud of color value data sample set and three-dimensional Gaussian cloud ambiguity degree and ambiguity degree threshold ratio compared with process, until the indistinct degree of all three-dimensional Gaussian clouds is respectively less than equal to described ambiguity degree threshold value.Wherein, described three-dimensional Gaussian cloud synthesis module includes three-dimensional Gaussian component synthon module and three-dimensional Gaussian cloud computing submodule.The color value data sample of described coloured image, according to quantity m of default initial three-dimensional Gaussian cloud, is converted into m three-dimensional Gaussian component G (< μ x by described three-dimensional Gaussian component synthon modulek,μyk,μzk>,σk), k=1 ..., m, wherein, the distribution that distributed constant is three primary colours expectation < the μ x of described m three-dimensional Gaussian componentk,μyk,μzk>, standard deviation sigmak, distribution weights.Described three-dimensional Gaussian cloud computing submodule utilizes described m three-dimensional Gaussian component G (< μ xk,μyk,μzk>,σk), k=1 ..., m, calculate the three-dimensional Gaussian YUNSHEN number that each three-dimensional Gaussian component is corresponding, described three-dimensional Gaussian YUNSHEN number includes the cloud expectation < R_Ex of three primary coloursk,G_Exk,B_Exk>, entropy Enk, super entropy Hek, ambiguity degree CDk, thus obtain m three-dimensional Gaussian cloud C (< R_Exk,G_Exk,B_Exk>,Enk,Hek),k=1,…,m。
Described three-dimensional Gaussian cloud output module is respectively less than all of three-dimensional Gaussian cloud equal to described ambiguity degree threshold value for exporting its ambiguity degree.
Described coloured image, for utilizing all of three-dimensional Gaussian cloud exported, is split by described color images module.
For piece image, described color value extraction module extracts the three primary colours color value of all pixels in image, forms color value data sample set, utilizes described three-dimensional Gaussian cloud synthesis module to generate m three-dimensional Gaussian cloud.Described three-dimensional Gaussian cloud synthesis module determines whether in the way of quantity m gradually subtracts one according to the indistinct degree of m three-dimensional Gaussian cloud, repeat the synthesis of three-dimensional Gaussian cloud, and three-dimensional Gaussian cloud ambiguity degree and ambiguity degree threshold value are compared process, final output meets multiple three-dimensional Gaussian clouds of condition, utilize the described multiple three-dimensional Gaussian clouds meeting condition for described color images module, carry out image segmentation or obtain the uncertain edge of transitional region in image.
In sum, the color value data sample set of any one coloured image can be converted into multiple Gauss cloud by the present invention, thus the uncertain edge of picture engraving transitional region, the fuzzy partitioning of feasible region, the prior indistinct degree being intended to utilize entropy and super entropy to be formed, come optimization gauss cloud quantity, granularity and level, solve the overlapping chaotic problem in image transition zone.
Although being described in detail the present invention above, but the invention is not restricted to this, those skilled in the art of the present technique can carry out various amendment according to the principle of the present invention.Therefore, all amendments made according to the principle of the invention, all should be understood to fall into protection scope of the present invention.
Claims (6)
1. a color image processing method based on three-dimensional Gaussian Cloud transform, it is characterised in that including:
A) by extracting the three primary colours color value of each pixel of coloured image, the color value of described coloured image is formed
Set of data samples；
B) according to default initial three-dimensional Gaussian cloud quantity m, by described color value data sample set synthesis m three
Dimension Gauss cloud；
C) successively the indistinct degree of m three-dimensional Gaussian cloud is compared with ambiguity degree threshold value, to determine whether there is
Its ambiguity degree is more than the three-dimensional Gaussian cloud of ambiguity degree threshold value；
D) if there is its ambiguity degree more than the three-dimensional Gaussian cloud of ambiguity degree threshold value, then according to by described quantity m by
The secondary mode subtracting, the process and the three-dimensional Gaussian cloud that repeat color value data sample set synthesis three-dimensional Gaussian cloud are indistinct
Degree with ambiguity degree threshold ratio compared with process, until the indistinct degree of all three-dimensional Gaussian clouds be respectively less than be equal to described ambiguity
Degree threshold value；
E) export its ambiguity degree and be respectively less than all of three-dimensional Gaussian cloud equal to described ambiguity degree threshold value；
Wherein, the indistinct degree of each three-dimensional Gaussian cloud is N times of super entropy and the entropy ratio of each three-dimensional Gaussian cloud,
N is positive integer；
Wherein, described step B) including:
B1) according to quantity m of default initial three-dimensional Gaussian cloud, by the color value data sample of described coloured image
Originally m three-dimensional Gaussian component it is converted into；
B2) utilize described m three-dimensional Gaussian component, calculate the three-dimensional Gaussian cloud that each three-dimensional Gaussian component is corresponding
Parameter, thus obtain m three-dimensional Gaussian cloud, including:
B21) pantograph ratio of m standard deviation of m three-dimensional Gaussian component is calculated respectively；
B22) utilize pantograph ratio and the distributed constant of described m three-dimensional Gaussian component, determine that each three-dimensional Gaussian divides
The three-dimensional Gaussian YUNSHEN number comprising the cloud expectation of three primary colours, entropy, super entropy, ambiguity degree that amount is corresponding, thus obtain
M three-dimensional Gaussian cloud.
Method the most according to claim 1, it is characterised in that described step B1) including:
B11) frequency distribution of three primary colours in described color value data sample set is added up；
B12) according to described quantity m, set respectively and include that the initial expectation of three primary colours, primary standard are poor, initial
The m group initial parameter of weights, and utilize described m group initial parameter, determine include three primary colours estimation expectation,
Estimated standard deviation, the m group of estimation weights estimate parameter；
B13) utilize the frequency distribution of described three primary colours and described m group initial parameter, calculate goal-selling function
Initial function value, and utilize the frequency distribution of described three primary colours and described m group to estimate parameter, calculate goal-selling
The estimation function value of function；
B14) difference between described estimation function value and described initial function value is calculated, if described difference is more than
In predictive error threshold value, then use described estimation parameter to replace described initial parameter, and repeat step B13),
Until described difference is less than predictive error threshold value；
B15) export described difference and estimate parameter less than the m group of predictive error threshold value, and described m group is estimated ginseng
Number is respectively as the m group distributed constant of m three-dimensional Gaussian component, wherein, estimates estimating of the three primary colours in parameter
Meter expectation, estimated standard deviation, estimation weights are respectively as the distribution expectation of the three primary colours in distributed constant, standard
Difference, distribution weights.
Method the most according to claim 2, it is characterised in that for described m three-dimensional Gaussian component
In kth three-dimensional Gaussian component, the calculation procedure of the pantograph ratio of its standard deviation includes:
In described m three-dimensional Gaussian component, three-dimensional with other by calculating kth three-dimensional Gaussian component respectively
The overlapping degree that Gaussian component is overlapping, obtains 1st maximum with kth three-dimensional Gaussian component degree of overlapping three-dimensional high
This component；
Described kth three-dimensional Gaussian component and described 1st three-dimensional Gaussian component are carried out the contracting of equal proportion
Put, obtain making the described kth three-dimensional Gaussian component scaled not overlap with described 1st three-dimensional Gaussian component
Pantograph ratio.
4. according to the method described in claim 1-3 any one, it is characterised in that also include:
Utilize all of three-dimensional Gaussian cloud exported, described coloured image is split.
5. a color image processing apparatus based on three-dimensional Gaussian Cloud transform, it is characterised in that including:
Color value extraction module, for the three primary colours color value by extracting each pixel of coloured image, forms institute
State the color value data sample set of coloured image；
Three-dimensional Gaussian cloud synthesis module, for according to default initial three-dimensional Gaussian cloud quantity m, by described color
Value Data sample set m three-dimensional Gaussian cloud of synthesis, and successively by the indistinct degree of m three-dimensional Gaussian cloud and ambiguity degree
Threshold value compares, to determine whether there is its ambiguity degree three-dimensional Gaussian cloud more than ambiguity degree threshold value, if existing
Its ambiguity degree is more than the three-dimensional Gaussian cloud of ambiguity degree threshold value, then according to the mode that described quantity m is gradually subtracted,
Repeat the synthesis process of three-dimensional Gaussian cloud of color value data sample set and three-dimensional Gaussian cloud ambiguity degree and ambiguity degree threshold
The process that value compares, until the indistinct degree of all three-dimensional Gaussian clouds is respectively less than equal to described ambiguity degree threshold value；
Three-dimensional Gaussian cloud output module, is respectively less than owning equal to described ambiguity degree threshold value for exporting its ambiguity degree
Three-dimensional Gaussian cloud；
Wherein, the indistinct degree of each three-dimensional Gaussian cloud is N times of super entropy and the entropy ratio of each three-dimensional Gaussian cloud,
N is positive integer；
Described three-dimensional Gaussian cloud synthesis module includes:
Three-dimensional Gaussian component synthon module, for quantity m according to default initial three-dimensional Gaussian cloud, by institute
The color value data sample stating coloured image is converted into m three-dimensional Gaussian component；
Three-dimensional Gaussian cloud computing submodule, is used for utilizing described m three-dimensional Gaussian component, calculates each three-dimensional high
The three-dimensional Gaussian YUNSHEN number that this component is corresponding, thus obtain m three-dimensional Gaussian cloud, including:
Calculate the pantograph ratio of m standard deviation of m three-dimensional Gaussian component respectively；
Utilize pantograph ratio and the distributed constant of described m three-dimensional Gaussian component, determine each three-dimensional Gaussian
What component was corresponding comprises the three-dimensional Gaussian YUNSHEN number of the cloud expectation of three primary colours, entropy, super entropy, ambiguity degree, thus obtains
To m three-dimensional Gaussian cloud.
Device the most according to claim 5, it is characterised in that also include:
Color images module, for utilizing all of three-dimensional Gaussian cloud exported, to described coloured image
Split.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310309949.5A CN103390276B (en) | 2013-07-23 | 2013-07-23 | A kind of color image processing method based on three-dimensional Gaussian Cloud transform and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310309949.5A CN103390276B (en) | 2013-07-23 | 2013-07-23 | A kind of color image processing method based on three-dimensional Gaussian Cloud transform and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103390276A CN103390276A (en) | 2013-11-13 |
| CN103390276B true CN103390276B (en) | 2016-09-28 |
Family
ID=49534535
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310309949.5A Expired - Fee Related CN103390276B (en) | 2013-07-23 | 2013-07-23 | A kind of color image processing method based on three-dimensional Gaussian Cloud transform and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103390276B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106251351B (en) * | 2016-07-28 | 2018-07-17 | 国网湖南省电力公司 | A kind of transmission line forest fire monitoring threshold computational methods based on Cloud transform |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101727657A (en) * | 2008-10-31 | 2010-06-09 | 李德毅 | Image segmentation method based on cloud model |
-
2013
- 2013-07-23 CN CN201310309949.5A patent/CN103390276B/en not_active Expired - Fee Related
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101727657A (en) * | 2008-10-31 | 2010-06-09 | 李德毅 | Image segmentation method based on cloud model |
Non-Patent Citations (2)
| Title |
|---|
| Granular computing based on Gaussian cloud transformation;Liu Yuchao 等;《Fundamenta Informaticae》;20130131;第127卷(第1-4期);385-398 * |
| 利用云模型和数据场的图像分割方法;吴涛 等;《模式识别与人工智能》;20120630;第25卷(第3期);397-405 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103390276A (en) | 2013-11-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106296695B (en) | Adaptive threshold natural target image segmentation extraction algorithm based on conspicuousness | |
| CN103914699A (en) | Automatic lip gloss image enhancement method based on color space | |
| CN107220949A (en) | The self adaptive elimination method of moving vehicle shade in highway monitoring video | |
| Machairas et al. | Waterpixels: Superpixels based on the watershed transformation | |
| CN109712165A (en) | A kind of similar foreground picture image set dividing method based on convolutional neural networks | |
| Lei et al. | Pix2surf: Learning parametric 3d surface models of objects from images | |
| CN114092697B (en) | Building facade semantic segmentation method with attention fused with global and local depth features | |
| CN104156997A (en) | Quick volume data skeleton extraction method based on rendering | |
| CN106991686A (en) | A kind of level set contour tracing method based on super-pixel optical flow field | |
| CN103177450A (en) | Image scene segmentation and layering joint solution method based on component set sampling | |
| CN105389821B (en) | It is a kind of that the medical image cutting method being combined is cut based on cloud model and figure | |
| CN103826032A (en) | Depth map post-processing method | |
| Nayak et al. | A modified triangle box-counting with precision in error fit | |
| CN103530866B (en) | A kind of image processing method based on Gaussian cloud transformation and device | |
| CN105303546A (en) | Affinity propagation clustering image segmentation method based on fuzzy connectedness | |
| CN102073867A (en) | Sorting method and device for remote sensing images | |
| CN109325451A (en) | A kind of high spatial resolution arable land full-automatic extractive technique in plot based on deep learning | |
| CN103729873B (en) | A kind of ambient light method of sampling of perception of content | |
| CN105913451B (en) | A kind of natural image superpixel segmentation method based on graph model | |
| CN103390276B (en) | A kind of color image processing method based on three-dimensional Gaussian Cloud transform and device | |
| CN104636728B (en) | A kind of image processing method | |
| CN105023288A (en) | A method for eliminating visual errors of two-dimensional vector solid lines in a three-dimensional scene | |
| CN107452001A (en) | A kind of remote sensing images sequences segmentation method based on improved FCM algorithm | |
| CN116958367A (en) | Method for quickly combining and rendering complex nerve scene | |
| CN102708554B (en) | Method for converting digital color images to grayscale images |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20160928 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |