CN103530866A - Image processing method and device based on Gaussian cloud transformation - Google Patents

Abstract

The invention discloses an image processing method and device based on Gaussian cloud transformation. The method comprises the steps of: forming a gray value data sample set of an image by exacting the gray value of each pixel; by counting the crest number of the frequency distribution of the gray value data sample set, obtaining a number m for forming an initial Gaussian cloud; composing the gray value data sample set to form m Gaussian clouds according to the number m; comparing the mixed degrees of the m Gaussian clouds with a mixed degree threshold to determine whether a Gaussian cloud, of which the mixed degree is larger than the mixed degree threshold, exists; if the Gaussian cloud, of which the mixed degree is larger than the mixed degree threshold, exists, repeating the processes of composing the gray value data sample set to form the Gaussian clouds in a mode of the number m minus one each time and comparing the mixed degrees of the Gaussian clouds with the mixed degree threshold until the mixed degrees of the Gaussian clouds are all less than or equal to the mixed degree threshold; and outputting all the Gaussian clouds of which the mixed degrees are less than or equal to the mixed degree threshold.

Description

A kind of image processing method and device based on the conversion of Gauss's cloud

Technical field

The present invention relates to image processing field, method and relevant apparatus thereof that particularly image processing is carried out in a kind of Gauss's of utilization cloud conversion.

Background technology

Grain calculates be research and simulating human from different grain size, different levels to things represent, the method for analysis and reasoning, be the important directions that in artificial intelligence, intelligent information processing technology is studied.In human cognitive, the selection of the expression of many granularities concept, generation and suitable particle size is that grain calculates the difficult problem facing always.

Cloud model is one and take the cognitive model that probability theory is fundamental research qualitative, quantitative conversion.Three numerical characteristic expectations for cloud model, entropy and super entropy characterize the intension of a qualitativing concept, be contemplated to be the data sample that can represent concept, entropy is the tolerance of concept granularity, the data area that reflection concept contains, super entropy is the uncertainty measure of concept granularity, the uncertainty on reflection data area border.

Gaussian distribution is most important distribution in theory of probability, the concept that relies on data statistics and take out usually has the feature of " broad in the middle, two is little ", standard deviation in Gaussian distribution can be used for characterizing the granularity of concept, the uncertainty of granularity that the super entropy of cloud model introducing has been weighed concept, be the uncertainty of standard deviation, so Gauss's cloud model is the most frequently used a kind of cloud model.

Cloud model is realized the conversion between a concept connotation and extension by forward cloud algorithm and reverse cloud algorithm.Reverse cloud algorithm can be converted to one group of data sample three numerical characteristics of a key concept, but the prerequisite of this algorithm acquiescence is, corresponding to same concept, the extension in same granularity characterizes given all data samples, and can not in whole Problem Areas, solve the Generating Problems of many granularities, many concepts.

In image is cut apart, image can be divided into ，Ru background area, several target area and foreground area according to the gray-scale value of pixel.At present, because Gaussian transformation is the size of the probable value in each Gaussian distribution according to each pixel correspondence in image, determine which target area pixel belongs to, thereby when image is cut apart, intersection in two regions usually there will be factitious sawtooth, can not embody the range of indeterminacy between target area; And the number of Gaussian transformation needs target area given in advance, cannot obtain adaptively image object quantity according to the statistical property of pixel gray-scale value.

Summary of the invention

For addressing the above problem, the object of the present invention is to provide a kind of image processing method and device based on the conversion of Gauss's cloud.

According to an aspect of the present invention, provide a kind of image processing method based on the conversion of Gauss's cloud, it is characterized in that, having comprised:

A) by extracting the gray-scale value of each pixel of image, form the gray value data sample set of described image;

B), by the crest quantity of the described gray value data sample set of statistics frequency distribution, obtain being used to form the quantity m of initial Gaussian cloud;

C) according to described quantity m, by synthetic m the Gauss's cloud of described gray value data sample set;

D) successively the indistinct degree of m Gauss's cloud and indistinct degree threshold value are compared, to determine whether existing its indistinct degree to be greater than Gauss's cloud of indistinct degree threshold value;

E) if exist its indistinct degree to be greater than Gauss's cloud of indistinct degree threshold value, according to the mode that described quantity m is successively subtracted to, repeat the synthetic processing of Gauss's cloud of gray value data sample set and the processing of Gauss's cloud ambiguity degree and the comparison of indistinct degree threshold value, until the indistinct degree of all Gauss's clouds is all less than or equal to described indistinct degree threshold value;

F) export all Gauss's clouds that its indistinct degree is all less than or equal to described indistinct degree threshold value.

Preferably, described step C) comprising:

C1) according to described quantity m, by synthetic m the Gaussian distribution of described gray value data sample set;

C2) utilize a described m Gaussian distribution, calculate Gauss's cloud parameter corresponding to each Gaussian distribution, obtain m Gauss's cloud.

Preferably, described step C1) comprising:

C11) add up the frequency distribution of described gray value data sample set;

C12), according to described quantity m, set respectively m and comprise that initial expectation, primary standard are poor, the initial parameter of initial weight;

C13) Offered target function, and utilize the frequency distribution of described gray value data sample set and a described m initial parameter, calculate respectively the initial function value of m objective function;

C14) utilize maximum likelihood to estimate and described initial parameter, calculate m and comprise the estimated parameter of estimating expectation, estimated standard deviation, estimation weights;

C15) utilize the frequency distribution of described gray value data sample set and a described m estimated parameter, calculate respectively the estimation function value of m objective function;

C16) 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, use described estimated parameter to replace described initial parameter, and repeating step C13) to step C15), until described difference is less than predictive error threshold value;

C17) export m the estimated parameter that described difference is less than predictive error threshold value, and using the estimation expectation of a described m estimated parameter its each, estimated standard deviation, estimated amplitude respectively as distribution expectation, standard value, the distribution weights of m its each distribution parameter of Gaussian distribution.

Preferably, described step C2) comprising:

C21) calculate respectively the pantograph ratio α of m standard deviation of m Gaussian distribution _k, k=1,2 ..., m;

C22) utilize the pantograph ratio α of a described m standard deviation _kwith a described m distribution parameter, calculate respectively m and comprise cloud expectation Ex _k, entropy En _k, super entropy He _k, indistinct degree CD _kgauss's cloud parameter, obtain m Gauss's cloud.

Preferably, the calculation procedure of the pantograph ratio of described m its each standard deviation of Gaussian distribution comprises:

By calculating the overlapping part of the weak outer peripheral areas of described Gaussian distribution left adjacent Gaussian distribution with it, obtain for making the first not overlapping pantograph ratio of this region;

By calculating the overlapping part of the weak outer peripheral areas of described Gaussian distribution right adjacent Gaussian distribution with it, obtain for making the second not overlapping pantograph ratio of this region;

Described the first pantograph ratio and described the second pantograph ratio are compared, and using smaller value wherein as the pantograph ratio α of the standard deviation of described Gaussian distribution _k.

Preferably, the weak outer peripheral areas of described Gaussian distribution left adjacent Gaussian distribution with it is [Ex _k-3En _k, Ex _k-2En _k], the weak outer peripheral areas of described Gaussian distribution right adjacent Gaussian distribution with it is [Ex _k+ 2En _k, Ex _k+ 3En _k].

Preferably, described indistinct degree is super entropy He _kwith entropy En _kdoubly, N is positive integer to the N of ratio.

Preferably, also comprise:

Utilize all Gauss's clouds of exporting, to described Image Segmentation Using.

According to a further aspect in the invention, provide a kind of image processing apparatus based on the conversion of Gauss's cloud, having comprised:

Gray-scale value extraction module, for by extracting the gray-scale value of each pixel of image, forms the gray value data sample set of described image;

Statistical module, for by the crest quantity of the described gray value data sample set of statistics frequency distribution, obtains being used to form the quantity m of Gauss's cloud;

Gauss's cloud synthesis module, be used for according to described quantity m, by synthetic m the Gauss's cloud of described gray value data sample set, successively the indistinct degree of m Gauss's cloud and indistinct degree threshold value are compared, to determine whether existing its indistinct degree to be greater than Gauss's cloud of indistinct degree threshold value, if exist its indistinct degree to be greater than Gauss's cloud of indistinct degree threshold value, according to the mode that described quantity m is successively subtracted to, repeat the synthetic processing of Gauss's cloud of gray value data sample set and the processing of Gauss's cloud ambiguity degree and the comparison of indistinct degree threshold value, until the indistinct degree of all Gauss's clouds is all less than or equal to described indistinct degree threshold value,

Gauss's cloud output module, is all less than or equal to all Gauss's clouds of described indistinct degree threshold value for exporting its indistinct degree.

Preferably, also comprise:

Image is cut apart module, for utilizing exported all Gauss's clouds, to described Image Segmentation Using.

Compared with prior art, beneficial effect of the present invention is:

The present invention extracts by the gray value data sample set of image being carried out to the conversion of Gauss's cloud and Gauss's cloud, synthetic suitable a plurality of Gauss's clouds, and utilize resulting a plurality of Gauss's cloud effectively to carry out image to cut apart, aspect the transitional region of image, uncertain rim detection, obtaining better segmentation effect.

Accompanying drawing explanation

Fig. 1 is a kind of image processing method process flow diagram based on the conversion of Gauss's cloud that the embodiment of the present invention provides;

Fig. 2 a is the former figure of grey value profile of certain each pixel of image of providing of the embodiment of the present invention;

Fig. 2 b is by the schematic diagram of synthetic two the Gauss's clouds of the grey value profile of Fig. 2 a by the conversion of Gauss's cloud;

Fig. 2 c is the conversion process schematic diagram of synthetic two Gauss's clouds in Fig. 2 b;

Fig. 3 a is the former figure that the embodiment of the present invention provides;

Fig. 3 b is the gray-scale value histogram of Fig. 3 a;

Fig. 3 c is three Gauss's clouds that form by the conversion of Gauss's cloud;

Fig. 3 d is Gauss's cloud of utilize generating result schematic diagram to Image Segmentation Using;

Fig. 3 e is the uncertain edge schematic diagram between black background and grey transitional region;

Fig. 3 f is the uncertain edge schematic diagram between grey transitional region and white laser region;

Fig. 4 is a kind of image processing apparatus block diagram based on the conversion of Gauss's cloud that the embodiment of the present invention provides.

Embodiment

Below in conjunction with accompanying drawing, to a preferred embodiment of the present invention will be described in detail, should be appreciated that following illustrated preferred embodiment, only for description and interpretation the present invention, is not intended to limit the present invention.

The invention discloses a kind of image processing method based on the conversion of Gauss's cloud, comprising: define indistinct degree, the gaussian component in gauss hybrid models GMM is converted to Gauss's cloud one by one with different meanings; According to indistinct degree, formulate Gauss's cloud and vary one's tactics, the quantity of optimization gauss cloud, and carry out image and cut apart.Experiment shows, with respect to prior art, the present invention has better segmentation effect at the transitional region of image, uncertain side edge detection mask.

The present invention includes from Gaussian transformation to Gauss's cloud vary one's tactics formulations, the extraction of Gauss's cloud and image of conversion, Gauss's cloud and the gordian technique such as cut apart.Wherein:

1, from Gaussian transformation to Gauss's cloud, convert

Any two the crossing Gaussian distribution that generate for Gaussian transformation, using the maximum particle size parameter of their current standard deviations as Gauss's cloud, according to " between class a little less than association, associated tight in class " principle, keep their expectation constant, carry out equal proportion reduction, until the element of their weak outer peripheral areas is non-intersect, now can obtain the minimum particle size parameter of each Gauss's cloud, utilize Gauss's cloud particle degree variation range can obtain entropy En and the super entropy He of each Gauss's cloud, owing to carrying out equal proportion reduction, so their He/En is identical, be called indistinct degree, its computing formula is as follows:

CD=N×He/En

Wherein, N is positive integer, and preferred, N value is 3.

Described indistinct degree can be used to weigh the dispersion degree of Gauss's cloud extension, is also the degree that the distribution of Gauss's cloud departs from Gaussian distribution.Ambiguity degree is 0, and Gauss's cloud extension converges, and builds consensus, and is ripe Gauss's cloud; Ambiguity degree is 1, and Gauss's cloud extension is dispersed, and is difficult to build consensus, i.e. atomization.

2, Gauss's cloud varies one's tactics

Preset indistinct degree threshold value, and utilize described indistinct degree threshold value formulation Gauss cloud to vary one's tactics, recursive call gauss hybrid models GMM and expectation maximization algorithm, meet until obtain Gauss's cloud quantity and the Gauss's cloud parameter thereof that indistinct degree threshold value requires.

Specifically, according to the overlapping degree of extension element between Gauss's cloud, can be divided into that key element is overlapping, fundamental element is overlapping, peripheral element is overlapping and weak peripheral element is overlapping.The chart that utilization ambiguity degree as shown in table 1 is measured Gaussian transformation division result.As can be seen from Table 1, Gauss's cloud, if indistinct degree is larger, outer postpone a meeting or conference more discrete, with the overlapping of adjacent Gauss's cloud conventionally can be more, therefore, more difficult building consensus; Otherwise indistinct degree is less, the extension of Gauss's cloud more converges, with the overlapping of adjacent Gauss's cloud conventionally can be fewer, therefore, more can build consensus.

Table 1

3, the extraction of Gauss's cloud and the image based on the conversion of Gauss's cloud cut apart

The conversion of Gauss's cloud adopts forward Gauss cloud algorithm to generate the degree of certainty of each pixel to Gauss's cloud, by compared pixels, put the degree of certainty size to different Gauss's clouds, judge which Gauss's cloud it belongs to, the degree of certainty that this calculating obtains has uncertainty, therefore can characterize the range of indeterminacy between adjacent Gauss's cloud.

By Fig. 1 to Fig. 4, the present invention is carried out to deep explanation below.

Fig. 1 is a kind of image processing method process flow diagram based on the conversion of Gauss's cloud that the embodiment of the present invention provides, and as shown in Figure 1, step comprises:.

Step 101, by extracting the gray-scale value of each pixel of image, form the gray value data sample set X{x of described image _i| i=1,2 ..., N}.

Step 102, by statistics described gray value data sample set frequency distribution p (x _i) crest quantity m, obtain being used to form the quantity m of initial Gaussian cloud, add up the crest quantity m that the frequency of gray value data sample set distributes, and the initial value using m as Gauss's cloud quantity.

Step 103, according to described quantity m, by synthetic m the Gauss's cloud of described gray value data sample set, specifically, utilize heuristic Gaussian transformation cloud (H_GCT) by gray value data sample set X{x _i| i=1,2 ..., N} is clustered into m Gauss's cloud C (Ex _k, En _k, He _k), k=1 ..., m.

Wherein, utilize heuristic Gaussian transformation cloud (H_GCT) that the step of the synthetic a plurality of Gauss's clouds of described gray value data sample set is comprised:

Step 103-1, add up described gray value data sample set X{x _i| i=1,2 ..., the frequency histogram of N}, i.e. frequency distribution p (x _i), and utilize Gaussian transformation to convert thereof into m Gaussian distribution:

G（μ _k,σ _k）|k=1,…,m

Step 103-2, for k Gaussian distribution, calculate the pantograph ratio α of its standard deviation _k, Gauss's cloud parameter of k Gauss's cloud is:

Ex _k=μ _k

En _k=(1+α _k)×σ _k/2

He _k=(1-α _k)×σ _k/6

CD _k=(1-α _k)/(1+α _k)

Further, described step 103-1 comprises:

Step 103-1-1, add up described gray value data sample set X{x _i| i=1,2 ..., the frequency distribution of N}.

h(y _i)=p(x _i),i＝1,2,…,N;j=1,2,…,N'

Wherein, y is sample domain space.

The initial parameter of step 103-1-2, m Gaussian distribution of setting, k (k=1 ..., m) initial parameter of individual Gaussian distribution is set as:

${\mathrm{\μ}}_k=\frac{k*\max \left(X\right)}{m+1}$

σ _k=max(X)

$a_k=\frac{1}{m}$

Step 103-1-3, objective definition function, and calculate initial function value

$J\left(\mathrm{\θ}\right)=\underset{i=1}{\overset{N^{\′}}{\mathrm{\Σ}}}[h\left(y_i\right)\×\ln \underset{k=1}{\overset{m}{\mathrm{\Σ}}}[a_{k}g(y_i;{\mathrm{\μ}}_k,{\mathrm{\σ}}_k^2)\left]\right]$

Wherein:

$g(y_i;{\mathrm{\μ}}_k,{\mathrm{\σ}}_k^2)=\frac{1}{\sqrt{2\mathrm{\π}}{\mathrm{\σ}}_k}{e}^{-\frac{{(y_i-{\mathrm{\μ}}_k)}^2}{2{\mathrm{\σ}}_k^2}}$

Step 103-1-4, to k (k=1 ..., m) individual Gaussian distribution, estimates according to maximum likelihood, calculates the estimated parameter of this Gaussian distribution

${\mathrm{\μ}}_k=\frac{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{L}_k\left(x_i\right)x_i}{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{L}_k\left(x_i\right)}$

${\mathrm{\σ}}_k^2=\frac{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{L}_k\left(x_i\right){(x_i-{\mathrm{\μ}}_k)}^T(x_i-{\mathrm{\μ}}_k)}{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{L}_k\left(x_i\right)}$

$a_k=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{L}_k\left(x_i\right)$

Wherein:

$L_k\left(x_i\right)=\frac{a_{k}g(x_i;{\mathrm{\μ}}_k,{\mathrm{\σ}}_k^2)}{\underset{n=1}{\overset{m}{\mathrm{\Σ}}}(a_{n}g(x_i;{\mathrm{\μ}}_n,{\mathrm{\σ}}_n^2)}$

The estimation function value of step 103-1-5, calculating target function.

$J\left(\widehat{\mathrm{\θ}}\right)=\underset{i=1}{\overset{N^{\′}}{\mathrm{\Σ}}}[h\left(y_i\right)\×\ln \underset{k=1}{\overset{m}{\mathrm{\Σ}}}[a_{k}g(y_i;{\mathrm{\μ}}_k,{\mathrm{\σ}}_k^2)\left]\right]$

Difference between step 103-1-6, the judgement estimation function value of objective function and the initial function value of objective function, if export current estimated parameter, otherwise adjust initial parameter, use current estimated parameter to upgrade initial parameter, and repeated execution of steps 103-1-3.

Step 104, successively the indistinct degree of m Gauss's cloud and indistinct degree threshold value are compared, to determine whether existing its indistinct degree to be greater than Gauss's cloud of indistinct degree threshold value.

If step 105 exists its indistinct degree to be greater than Gauss's cloud of indistinct degree threshold value, according to the mode that described quantity m is successively subtracted to, repeat the synthetic processing of Gauss's cloud of gray value data sample set and the processing of Gauss's cloud ambiguity degree and the comparison of indistinct degree threshold value, until the indistinct degree of all Gauss's clouds is all less than or equal to described indistinct degree threshold value.

Step 106, export all Gauss's clouds that its indistinct degree is all less than or equal to described indistinct degree threshold value.

In described step 104, to step 106, by indistinct degree order from large to small, m Gauss's cloud sorted, and successively the indistinct degree CD of each Gauss's cloud is judged, if CD _k>β, k=1 ..., m, β is indistinct degree threshold value given in advance, quantity m subtracts 1, and repeated execution of steps 2, otherwise, export all Gauss's cloud C (Ex that indistinct degree is less than or equal to β _k, En _k, He _k), k=1 ..., m'.

Wherein, by indistinct degree order, indistinct degree CD to each Gauss's cloud judges, comprise following content: for k Gaussian distribution in Gaussian transformation, calculate respectively the overlapping degree between its two Gaussian distribution adjacent with left and right, if their weak peripheral element, 2 σ are not overlapping to outer peripheral areas a little less than 3 σ, illustrate that Gauss's cloud that this Gaussian distribution represents divides clearly, its entropy is exactly the standard deviation (En in Gaussian distribution _k=σ _k), super entropy is 0(He _k=0); Otherwise, illustrate between this Gauss's cloud and adjacent Gauss's cloud and exist and divide unsharp overlapping region, keep their expectation values constant, their standard deviation is carried out to convergent-divergent by equal proportion, calculate and obtain the weak not overlapping scaling α of peripheral element between Gauss's cloud adjacent with left side ₁, meet

μ _k-1+3*a ₁*σ _k-1＝μ _k-3*a ₁*σ _k，

Calculate and obtain the weak nonoverlapping scaling α of peripheral element between Gauss's cloud adjacent with right side ₂, meet

μ _k+3*a ₂*σ _k=μ _k+1-3*a ₂*σ _k+1

The standard deviation variation range that k Gaussian distribution causes because conception division is unintelligible is [α * σ _k, σ _k], α=min (α ₁, α ₂).According to the definition of Gauss's cloud, entropy is the expectation of standard deviation, and super entropy is the standard deviation of standard deviation, meets equally 3 σ principles, therefore, can calculate and obtain the Gauss's cloud parameter that characterizes k concept:

Ex _k=μ _k

En _k=(1+α _k)×σ _k/2

He _k=(1-α _k)×σ _k/6

He _k/En _k=(1-α)/3(1+α)

Ambiguity degree is:

CD _k=3×He _k/En _k=(1-α)/(1+α)

By the order of indistinct degree, if maximum indistinct degree CD _k>β, k=1 ..., m, illustrates and has the overlapping situation of fundamental element between two concepts, concept is counted m=m-1.

Fig. 2 shown and utilizes Gauss's cloud transfer pair width image to carry out the result of cluster, and Fig. 2 a is the former figure of grey value profile of certain each pixel of image of providing of the embodiment of the present invention, i.e. gray-scale value probability distribution graph; Fig. 2 b is by the schematic diagram of synthetic two the Gauss's clouds of the grey value profile of Fig. 2 a by the conversion of Gauss's cloud, wherein, except original gray-scale value probability distribution graph, also comprise the matched curve representing with solid line, with the expectation curve of the Gauss's cloud dotting, described expectation curve; Fig. 2 c is the conversion process schematic diagram of synthetic two Gauss's clouds in Fig. 2 b, by Gauss's cloud conversion process, 5 Gauss's clouds is converted to the change granularity process of 2 Gauss's clouds.

Fig. 3 has shown that 256 * 256 pixels, gray-scale value are [0,255] the laser melting coating figure between and utilize Gauss's cloud transfer pair it carries out the result that gray-scale value extracts and image is cut apart, white portion is the laser of high-energy-density, black region is background color, has a transitional region simultaneously between prospect and background.Fig. 3 a is the former figure that the embodiment of the present invention provides, and by extracting the gray-scale value of each pixel of image, forms the gray value data sample set of described image, adds up the frequency distribution of described gray value data sample set, and obtaining Fig. 3 (b) is its grey level histogram.By adding up the crest quantity of described gray value data sample set frequency distribution, obtain being used to form the quantity m of Gauss's cloud.1), add up the frequency distribution of described gray value data sample set then carry out following steps:; 2), according to described quantity m, set respectively m and comprise that initial expectation, primary standard are poor, the initial parameter of initial weight; 3) Offered target function, and utilize the frequency distribution of described gray value data sample set and a described m initial parameter, calculate respectively the initial function value of m objective function; 4) utilize maximum likelihood to estimate and described initial parameter, calculate m and comprise the estimated parameter of estimating expectation, estimated standard deviation, estimation weights; 5) utilize the frequency distribution of described gray value data sample set and a described m estimated parameter, calculate respectively the estimation function value of m objective function; 6) 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, use described estimated parameter to replace described initial parameter, and repeating step 3) to step 5), until described difference is less than predictive error threshold value; 7) export m the estimated parameter that described difference is less than predictive error threshold value, and using the estimation expectation of a described m estimated parameter its each, estimated standard deviation, estimated amplitude respectively as distribution expectation, standard value, the distribution weights of m its each distribution parameter of Gaussian distribution.8) calculate respectively the pantograph ratio α of m standard deviation of m Gaussian distribution _k, k=1,2 ..., m; 9) utilize the pantograph ratio α of a described m standard deviation _kwith a described m distribution parameter, calculate respectively m and comprise cloud expectation Ex _k, entropy En _k, super entropy He _k, indistinct degree CD _kgauss's cloud parameter, obtain m Gauss's cloud; 10) successively the indistinct degree of m Gauss's cloud and indistinct degree threshold value are compared, to determine whether existing its indistinct degree to be greater than Gauss's cloud of indistinct degree threshold value; 11) if exist its indistinct degree to be greater than Gauss's cloud of indistinct degree threshold value, according to the mode that described quantity m is successively subtracted to, repeat the synthetic processing of Gauss's cloud of gray value data sample set and the processing of Gauss's cloud ambiguity degree and the comparison of indistinct degree threshold value, until the indistinct degree of all Gauss's clouds is all less than or equal to described indistinct degree threshold value; 12) export Gauss's cloud of 3 that its indistinct degree is all less than or equal to described indistinct degree threshold value, as shown in Figure 3 c.Utilize 3 Gauss's clouds that generate to Image Segmentation Using, obtain segmentation result as shown in Figure 3 d, overlapping owing to existing between Gauss's cloud, the uncertainty on this division border can be utilized the super entropy of Gauss's cloud to calculate and obtain, therefore the image based on the conversion of Gauss's cloud is cut apart the range of indeterminacy that can divide between concept, Fig. 3 e has shown the uncertain edge between black background and grey transitional region, and Fig. 3 f has shown the uncertain edge between grey transitional region and white laser region.

Fig. 4 is a kind of image processing apparatus block diagram based on the conversion of Gauss's cloud that the embodiment of the present invention provides, and as shown in Figure 4, comprising:

Gray-scale value extraction module, for by extracting the gray-scale value of each pixel of image, forms the gray value data sample set of described image;

Statistical module, for by the crest quantity of the described gray value data sample set of statistics frequency distribution, obtains being used to form the quantity m of Gauss's cloud;

Gauss's cloud synthesis module, be used for according to described quantity m, by synthetic m the Gauss's cloud of described gray value data sample set, successively the indistinct degree of m Gauss's cloud and indistinct degree threshold value are compared, to determine whether existing its indistinct degree to be greater than Gauss's cloud of indistinct degree threshold value, if exist its indistinct degree to be greater than Gauss's cloud of indistinct degree threshold value, according to the mode that described quantity m is successively subtracted to, repeat the synthetic processing of Gauss's cloud of gray value data sample set and the processing of Gauss's cloud ambiguity degree and the comparison of indistinct degree threshold value, until the indistinct degree of all Gauss's clouds is all less than or equal to described indistinct degree threshold value,

Gauss's cloud output module, is all less than or equal to all Gauss's clouds of described indistinct degree threshold value for exporting its indistinct degree;

Image is cut apart module, for utilizing exported all Gauss's clouds, to described Image Segmentation Using.

For piece image, gray-scale value extraction module extracts the gray-scale value of all pixels in image, form gray value data sample set, for the crest quantity of gray value data sample set frequency distribution described in statistical module counts, and using described quantity as the quantity m that is used to form Gauss's cloud.Gauss's cloud synthesis module is first according to described quantity m, by synthetic m the Gauss's cloud of described gray value data sample set, and determine whether successively to subtract with quantity m one mode according to the indistinct degree of m Gauss's cloud, repeat the synthetic processing of Gauss's cloud of gray value data sample set and the processing of Gauss's cloud ambiguity degree and the comparison of indistinct degree threshold value, and according to result, a plurality of Gauss's clouds that satisfied condition by the final output of Gauss's cloud output module, for image, cut apart module and utilize described a plurality of Gauss's cloud, carry out the uncertain edge that transitional region in image is cut apart or obtained to image.

The present invention can convert the gray value data sample set of any one image to a plurality of Gauss's clouds, thereby the uncertain edge of picture engraving transitional region, the soft division of feasible region, the more important thing is the indistinct degree that will utilize entropy and super entropy to form, come optimization gauss cloud quantity, granularity and level, solve the problem of the overlapping confusion in image transition region.

Although above the present invention is had been described in detail, the invention is not restricted to this, those skilled in the art of the present technique can carry out various modifications according to principle of the present invention.Therefore, all modifications of doing according to the principle of the invention, all should be understood to fall into protection scope of the present invention.

Claims (10)

1. the image processing method based on the conversion of Gauss's cloud, is characterized in that, comprising:

A) by extracting the gray-scale value of each pixel of image, form the gray value data sample set of described image;

B), by the crest quantity of the described gray value data sample set of statistics frequency distribution, obtain being used to form the quantity m of initial Gaussian cloud;

C) according to described quantity m, by synthetic m the Gauss's cloud of described gray value data sample set;

D) successively the indistinct degree of m Gauss's cloud and indistinct degree threshold value are compared, to determine whether existing its indistinct degree to be greater than Gauss's cloud of indistinct degree threshold value;

E) if exist its indistinct degree to be greater than Gauss's cloud of indistinct degree threshold value, according to the mode that described quantity m is successively subtracted to, repeat the synthetic processing of Gauss's cloud of gray value data sample set and the processing of Gauss's cloud ambiguity degree and the comparison of indistinct degree threshold value, until the indistinct degree of all Gauss's clouds is all less than or equal to described indistinct degree threshold value;

F) export all Gauss's clouds that its indistinct degree is all less than or equal to described indistinct degree threshold value.

2. method according to claim 1, is characterized in that, described step C) comprising:

C1) according to described quantity m, by synthetic m the Gaussian distribution of described gray value data sample set;

C2) utilize a described m Gaussian distribution, calculate Gauss's cloud parameter corresponding to each Gaussian distribution, obtain m Gauss's cloud.

3. method according to claim 2, is characterized in that, described step C1) comprising:

C11) add up the frequency distribution of described gray value data sample set;

C12), according to described quantity m, set respectively m and comprise that initial expectation, primary standard are poor, the initial parameter of initial weight;

C13) Offered target function, and utilize the frequency distribution of described gray value data sample set and a described m initial parameter, calculate respectively the initial function value of m objective function;

C14) utilize maximum likelihood to estimate and described initial parameter, calculate m and comprise the estimated parameter of estimating expectation, estimated standard deviation, estimation weights;

C15) utilize the frequency distribution of described gray value data sample set and a described m estimated parameter, calculate respectively the estimation function value of m objective function;

C16) 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, use described estimated parameter to replace described initial parameter, and repeating step C13) to step C15), until described difference is less than predictive error threshold value;

C17) export m the estimated parameter that described difference is less than predictive error threshold value, and using the estimation expectation of a described m estimated parameter its each, estimated standard deviation, estimated amplitude respectively as distribution expectation, standard value, the distribution weights of m its each distribution parameter of Gaussian distribution.

4. method according to claim 3, is characterized in that, described step C2) comprising:

C21) calculate respectively the pantograph ratio α of m standard deviation of m Gaussian distribution _k, k=1,2 ..., m;

C22) utilize the pantograph ratio α of a described m standard deviation _kwith a described m distribution parameter, calculate respectively m and comprise cloud expectation Ex _k, entropy En _k, super entropy He _k, indistinct degree CD _kgauss's cloud parameter, obtain m Gauss's cloud.

5. method according to claim 4, is characterized in that, the calculation procedure of the pantograph ratio of described m its each standard deviation of Gaussian distribution comprises:

By calculating the overlapping part of the weak outer peripheral areas of described Gaussian distribution left adjacent Gaussian distribution with it, obtain for making the first not overlapping pantograph ratio of this region;

By calculating the overlapping part of the weak outer peripheral areas of described Gaussian distribution right adjacent Gaussian distribution with it, obtain for making the second not overlapping pantograph ratio of this region;

Described the first pantograph ratio and described the second pantograph ratio are compared, and using smaller value wherein as the pantograph ratio α of the standard deviation of described Gaussian distribution _k.

6. method according to claim 5, is characterized in that, the weak outer peripheral areas of described Gaussian distribution left adjacent Gaussian distribution with it is [Ex _k-3En _k, Ex _k-2En _k], the weak outer peripheral areas of described Gaussian distribution right adjacent Gaussian distribution with it is [Ex _k+ 2En _k, Ex _k+ 3En _k].

7. method according to claim 4, is characterized in that, described indistinct degree is super entropy He _kwith entropy En _kdoubly, N is positive integer to the N of ratio.

8. according to the method described in claim 1-7 any one, it is characterized in that, also comprise:

Utilize all Gauss's clouds of exporting, to described Image Segmentation Using.

9. the image processing apparatus based on the conversion of Gauss's cloud, is characterized in that, comprising:

Gray-scale value extraction module, for by extracting the gray-scale value of each pixel of image, forms the gray value data sample set of described image;

Statistical module, for by the crest quantity of the described gray value data sample set of statistics frequency distribution, obtains being used to form the quantity m of initial Gaussian cloud;

Gauss's cloud synthesis module, be used for according to described quantity m, by synthetic m the Gauss's cloud of described gray value data sample set, successively the indistinct degree of m Gauss's cloud and indistinct degree threshold value are compared, to determine whether existing its indistinct degree to be greater than Gauss's cloud of indistinct degree threshold value, if exist its indistinct degree to be greater than Gauss's cloud of indistinct degree threshold value, according to the mode that described quantity m is successively subtracted to, repeat the synthetic processing of Gauss's cloud of gray value data sample set and the processing of Gauss's cloud ambiguity degree and the comparison of indistinct degree threshold value, until the indistinct degree of all Gauss's clouds is all less than or equal to described indistinct degree threshold value,

Gauss's cloud output module, is all less than or equal to all Gauss's clouds of described indistinct degree threshold value for exporting its indistinct degree.

10. method according to claim 9, is characterized in that, also comprises:

Image is cut apart module, for utilizing exported all Gauss's clouds, to described Image Segmentation Using.

Images