CN103400370A - Adaptive fuzzy C-means image segmentation method based on potential function - Google Patents

Abstract

The invention provides an adaptive fuzzy C-means image segmentation method based on a potential function and mainly aims at solving the problems that classifying number needs to be previously set and the segmentation efficiency is low in the existing fuzzy C-means segmentation method. The method is realized by the following steps: (1) inputting a to-be-segmented image; (2) obtaining the histogram potential function and the maximum potential remnant height of the to-be-segmented image; (3) obtaining the histogram c-factorial remnant potential function of the to-be-segmented image; (4) combining pseudo potentials; (5) obtaining the clustering center and the classifying number of the to-be-segmented image; (6) carrying out fuzzy classification on the pixel dots of the to-be-segmented image; and (7) outputting an image segmentation result. The adaptive fuzzy C-means image segmentation method has the advantages of capability of obtaining optimal image classifying number in an adaptive way, high segmentation efficiency, strong region consistency in the segmentation result, smooth margin and the like. The adaptive fuzzy C-means image segmentation method can effectively segment natural gray images and infrared images and can be used for target recognition and tracking.

Description

Adaptive fuzzy C average image partition method based on potential function

Technical field

The invention belongs to field of computer technology, further relate to the C of the adaptive fuzzy based on the potential function average image partition method in technical field of image processing.The present invention obtains cluster centre and the optimal classification number of image to be split by the potential function method, and to classify of image element and the mark of image, realization is cut apart natural gray image and infrared image, can be applicable to target recognition and tracking.

Background technology

Image segmentation can think the pixel in image is carried out the process of cluster, according to the gray feature of each pixel in image, judges which kind of this pixel is under the jurisdiction of, and, with being under the jurisdiction of of a sort pixel mark in image, completes cutting apart image.Due to the fuzzy partition method meet the human cognitive characteristic, describe brief introduction distinct, be easy to realize, and than traditional hard dividing method, can keep more original image information, this dividing method more and more causes people's concern.

Proposed by Dunn,, by the standard fuzzy C-mean algorithm dividing method that Bezdekl promotes, be widely used in the fields such as graphical analysis, medical diagnosis, target identification and image segmentation.The implementation procedure of standard fuzzy C-mean algorithm dividing method is: at first, and the degree of membership of classification number, cluster centre and each pixel of initialisation image; Secondly, by the interative computation method, cluster centre and the pixel degree of membership of image are upgraded; Finally, according to the degree of membership of each pixel in image, pixel is classified, will be under the jurisdiction of of a sort pixel mark, realize cutting apart image.The weak point of standard fuzzy C-mean algorithm dividing method is, at first, before to image segmentation, the method need to set in advance the classification number of image, and the optimal classification number of image often can not be determined in advance before image segmentation, therefore, the artificial Images Classification number that arranges can not make the segmentation effect of image reach optimum usually; Secondly, the method for standard fuzzy C-mean algorithm dividing method by interative computation be the cluster centre of new images more, and this process computational complexity is higher, causes the method very slow to the splitting speed of image, is unfavorable for realizing the Real-time segmentation to image.

Zhang Yongchang paper " in conjunction with the image partition method of fuzzy clustering algorithm " (" development of computers and application ", article numbering: 1003-5850 (2011) 11-0049-03).The implementation procedure of the method is, at first image to be split is carried out windowing process, neighborhood zone after the acquisition windowing, then with the gray average of pixel in the neighborhood zone and gray variance normalization, and with the gray average after normalization and gray variance as this regional gray feature, the sextuple proper vector of structure description characteristics of image, adopt the fuzzy C-mean algorithm dividing method to the region unit cluster in image finally, realizes cutting apart image.The weak point of the method is, although suppressing the impact of noise aspect the Feature Selection of image, but the fuzzy C-mean algorithm dividing method of following adopted, still need artificially to arrange the classification number of image, can not guarantee to make the segmentation effect of image to reach optimum, simultaneously, the method for fuzzy C-mean algorithm dividing method by interative computation be the cluster centre of new images more, the computing complicated and time consumption, be unfavorable for realizing the Real-time segmentation to image.

The patent " a kind of color image segmentation method " of Zhongxing Microelectronci Co., Ltd., Beijing's application (number of patent application: 200810055833.2, publication No.: CN101216890A).The implementation procedure of the method is, at first set the classification number of image to be split, then use multiple different fuzzy clustering algorithm respectively to this Image Segmentation Using, correspondingly obtain a plurality of degree of membership matrixes, and use one of them degree of membership matrix as Criterion-matrix, the classification mark of other degree of membership matrixes of registration, merge the degree of membership matrix after registration and Criterion-matrix, and according to the degree of membership matrix computations after merging, go out classification mark corresponding to each pixel, realize cutting apart image.the weak point of the method is, although the mode that the method adopts a plurality of degree of membership matrixes to merge, can obtain more exactly the classification mark of pixel in image, make segmentation effect better, but the classification number of the image to be split that prerequisite is the method to be set should be optimum, if the classification number of setting is improper, the method still can not obtain segmentation result preferably, and the method needs to calculate a plurality of degree of membership matrixes in calculating process, and degree of membership is carried out the registration fusion, computational complexity is very high, cause splitting speed slow, be unfavorable for realizing the Real-time segmentation to image.

Summary of the invention

The purpose of the method for the present invention is that overcomes the deficiency of above-mentioned prior art, a kind of C of adaptive fuzzy based on potential function average image partition method is proposed, obtain optimal classification number and the cluster centre of image to be split with potential function clustering method self-adaptation, replace in the fuzzy C-mean algorithm dividing method artificially arranging classification number and iteration renewal cluster centre, realize the accurate classification to image, improved segmentation effect and splitting speed.

Concrete steps of the present invention are as follows:

(1) input image to be split

(2) obtain image histogram potential function to be split and maximal potential residual altitude

2a) adopt the Normalized Grey Level statistic histogram of image to be split, calculate and obtain image histogram potential function to be split;

Step 2a) described image histogram potential function to be split calculates by following formula:

$P\left(k\right)={\mathrm{\Σ}}_{i=0}^{255}H\left(i\right)/(1+\mathrm{\α}{(i-k)}^2)$

Wherein, P (k) represents image histogram potential function to be split, k represents the number of greyscale levels in image histogram potential function to be split, span is 0≤k≤255, ∑ represents that H (i) represents image normalization gray-scale statistical histogram to be split to the expression formula summation, and i represents the number of greyscale levels in image normalization gray-scale statistical histogram to be split, span is 0≤i≤255, and α represents the Constant control factor.

2b) by calculating the image histogram potential function to be split that obtains, calculate according to the following formula the maximal potential residual altitude:

R＝min{P ₁，P ₂，…，P _n}

Wherein, R represents the maximal potential residual altitude, and span is 0＜R＜1, and min represents the expression formula in { } is minimized, P ₁, P ₂..., P _nThe peak value that represents each crest in image histogram potential function curve to be split, n represent the number of image histogram potential function curve medium wave peak to be split;

(3) obtain image histogram c to be split rank residue potential function

3a) the preliminary classification number of image to be split is set to 2;

3b) calculate fuzzy decay factor and the fuzzy pseudopotential factor;

Step 3b) described fuzzy decay factor calculates by following formula:

f ₁＝(L-1)/D ²

Step 3b) the described fuzzy pseudopotential factor calculates by following formula:

f ₂＝D/3L

Wherein, f ₁Represent fuzzy decay factor, f ₂Represent the fuzzy pseudopotential factor, L presentation class number, D represent the gradation of image degree of depth to be split, and its value is poor for gray scale maximal value and minimum value in image slices vegetarian refreshments to be split.

3c) in image histogram c to be split rank residue potential function group, the initial number of pseudopotential is set to 0;

3d) calculate image histogram c to be split rank residue potential function, and it is poor to calculate in the group of functions that obtains gesture extreme point corresponding to two adjacent in twos rank residue potential functions, when difference is arranged less than the value of the fuzzy pseudopotential factor, just make the pseudopotential number of group of functions add 1 in all differences that obtain;

Step 3d) residue potential function in described image histogram c to be split rank calculates by following formula:

$P_c\left(k\right)=P_{c-1}\left(k\right)-P_{c-1}^{*}/(1+f_1{(k-x_{c-1})}^4)$

Wherein, P _c(k) expression image histogram c to be split rank residue potential function, subscript c represents image histogram c to be split rank residue potential function, span is c=1,2 ..., L+T, L presentation class number, T represents pseudopotential number, P _c-1(k) expression image c-1 to be split rank residue potential function,

The maximal value that represents image c-1 to be split rank residue potential function, f ₁Represent fuzzy decay factor, k represents number of greyscale levels, and span is 0≤k≤255, x _c-1The gesture extreme point that represents image c-1 to be split rank residue potential function.

Whether the peak value that 3e) judges the last single order residue of image histogram to be split potential function is less than the maximal potential residual altitude, if peak value, less than the maximal potential residual altitude, performs step (4), otherwise order classification number adds 1, execution step 3b);

(4) merge pseudopotential

4a) compute histograms gesture partition function;

Step 4a) described histogram gesture partition function calculates by following formula:

$F_c\left(k\right)=P_{c-1}^{*}/(1+f_1{(k-x_{c-1})}^4)$

Wherein, F _c(k) expression histogram gesture partition function, subscript c represents c rank histogram gesture partition function, span is c=1,2 ..., L+T, L presentation class number, T represents the pseudopotential number, and k represents number of greyscale levels, and span is 0≤k≤255,

The maximal value that represents image c-1 to be split rank residue potential function, f ₁Represent fuzzy decay factor, x _c-1The gesture extreme point that represents image c-1 to be split rank residue potential function.

4b) will calculate all gesture extreme points corresponding to histogram gesture partition function group that obtain and sort according to ascending order, and according to the clooating sequence of gesture extreme point, corresponding with it histogram gesture partition function be arranged;

4c) judge in histogram gesture partition function group whether have pseudopotential

With in the histogram gesture partition function group after arranging in twos adjacent gesture extreme point corresponding to two gesture partition functions subtract each other, in all differences that obtain when occurring difference less than the value of the fuzzy pseudopotential factor, show in two corresponding with it histogram gesture partition functions and have pseudopotential, perform step 4d), if do not have pseudopotential in histogram gesture partition function group, with this group of functions as image histogram gesture partition function group to be split, the execution step (5);

4d) the pseudopotential in merging histogram gesture partition function group

Extract respectively the extreme value that has two histogram gesture partition functions of pseudopotential in histogram gesture partition function group, relatively two extreme value acquisitions maximal value wherein; , with there being two histogram gesture partition function summations of pseudopotential, obtain one and function; Will with function match according to the following formula, obtain merging the histogram gesture partition function after pseudopotential:

F′(k)＝y/(1+f ₁(k-x) ⁴)

Wherein, F ' (k) represents to merge histogram gesture partition function after pseudopotential, and k represents number of greyscale levels, and span is 0≤k≤255, and y represents to exist the maximal value of two histogram gesture partition functions of pseudopotential, f ₁Represent fuzzy decay factor, x represents the extreme point with function;

4e) all pseudopotentials in histogram gesture partition function group are merged, obtain image histogram gesture partition function group to be split;

(5) obtain cluster centre and the optimal classification number of image to be split

With each gesture extreme point corresponding to image histogram gesture partition function group to be split, as the cluster centre of image to be split, with the optimal classification number of the number of gesture extreme point as image to be split;

(6) with the pixel fuzzy classification of image to be split

6a) the Euclidean distance of each pixel and cluster centre in calculating image to be split;

Step 6a) in described image to be split, the Euclidean distance of pixel and cluster centre calculates by following formula:

D _ij＝(I _i-V _j) ²

Wherein, D _ijRepresent the Euclidean distance of i pixel of image to be split to j cluster centre, I _iThe gray-scale value that represents i pixel of image to be split, V _jJ the cluster centre that represents image to be split.

6b) by the degree of membership of calculating pixel in the Euclidean distance calculating image to be split that obtains;

Step 6b) in described image to be split, the degree of membership of pixel calculates by following formula:

$u_{\mathrm{ij}}=1/{\mathrm{\Σ}}_{n=1}^M{(D_{\mathrm{ij}}/D_{\mathrm{in}})}^2$

Wherein, u _ijRepresent in image to be split that i pixel is under the jurisdiction of the degree of membership of j cluster centre, ∑ represents that M represents the optimal classification number of image to be split to the expression formula summation, and n represents n class in the optimal classification number of image to be split, D _ijRepresent the Euclidean distance of i pixel of image to be split to j cluster centre, D _inRepresent the Euclidean distance of i pixel of image to be split to n cluster centre.

6c) by the degree of membership of the pixel that calculate to obtain to the classify of image element in image to be split;

(7) output image segmentation result

To be under the jurisdiction of of a sort pixel mark in image to be split, complete the cutting apart of image, and the output image segmentation result.

The present invention compared with prior art has following advantage:

The first, owing to adopting potential function clustering method self-adaptation to obtain the optimal classification number of image to be split in the inventive method, overcome prior art and needed artificially to arrange the classification number before image segmentation, when the classification number that arranges is not the optimal classification number of image to be split, cause the poor problem of image segmentation, make the inventive method can realize accurate classification to image having improved segmentation effect.

Description of drawings

Fig. 1 is process flow diagram of the present invention;

Fig. 2 is the simulated effect figure of the present invention to a width aircraft natural gray image and prior art;

Fig. 3 is the simulated effect figure of the present invention to a width helicopter infrared image and prior art;

Fig. 4 is the simulated effect figure of the present invention to armed people's infrared image and prior art.

Embodiment

Be described in further detail below in conjunction with 1 pair of specific embodiment of the invention step of accompanying drawing.

Step 1. input image to be split.In embodiments of the present invention,, by WINDOWS XP system input image to be split, obtain the intensity profile matrix of this image slices vegetarian refreshments.

Step 2. is obtained image histogram potential function to be split and maximal potential residual altitude

Adopt the Normalized Grey Level statistic histogram of image to be split, by following formula, calculate and obtain image histogram potential function to be split:

$P\left(k\right)={\mathrm{\Σ}}_{i=0}^{255}H\left(i\right)/(1+\mathrm{\α}{(i-k)}^2)$

Wherein, P (k) represents image histogram potential function to be split, k represents the number of greyscale levels in image histogram potential function to be split, span is 0≤k≤255, ∑ represents that H (i) represents image normalization gray-scale statistical histogram to be split to the expression formula summation, and i represents the number of greyscale levels in image normalization gray-scale statistical histogram to be split, span is 0≤i≤255, and α represents the Constant control factor.In embodiments of the present invention, α=0.05.

Calculate according to the following formula the maximal potential residual altitude by calculating the image histogram potential function to be split that obtains:

R＝min{P ₁，P ₂，…，P _n}

Wherein, R represents the maximal potential residual altitude, and span is 0＜R＜1, and min represents the expression formula in { } is minimized, P ₁, P ₂..., P _nThe peak value that represents each crest in image histogram potential function curve to be split, n represent the number of image histogram potential function curve medium wave peak to be split.

Step 3. is obtained image histogram c to be split rank residue potential function

3a) the preliminary classification number of image to be split is set to 2;

3b) by following formula, calculate and obtain fuzzy decay factor:

f ₁＝(k-1)/D ²

Calculate and obtain the fuzzy pseudopotential factor by following formula:

f ₂＝D/3L

Wherein, f ₁Represent fuzzy decay factor, f ₂Represent the fuzzy pseudopotential factor, L presentation class number, D represent the gradation of image degree of depth to be split, and its value is poor for gray scale maximal value and minimum value in image slices vegetarian refreshments to be split.

3c) in image histogram c to be split rank residue potential function group, the initial number of pseudopotential is set to 0;

3d) by following formula, calculate and obtain image histogram c to be split rank residue potential function:

$P_c\left(k\right)=P_{c-1}\left(k\right)-P_{c-1}^{*}/(1+f_1{(k-x_{c-1})}^4)$

Wherein, P _c(k) expression image histogram c to be split rank residue potential function, subscript c represents image histogram c to be split rank residue potential function, span is c=1,2 ..., L+T, L presentation class number, T represents pseudopotential number, P _c-1(k) expression image c-1 to be split rank residue potential function,

The maximal value that represents image c-1 to be split rank residue potential function, f ₁Represent fuzzy decay factor, k represents number of greyscale levels, and span is 0≤k≤255, x _c-1The gesture extreme point that represents image c-1 to be split rank residue potential function., with in the group of functions that calculate to obtain, gesture extreme point corresponding to adjacent two rank residue potential functions is poor in twos, when difference is arranged less than the value of the fuzzy pseudopotential factor, just make the pseudopotential number of group of functions add 1 in all differences of acquisition;

Whether the peak value that 3e) judges the last single order residue of image histogram to be split potential function is less than the maximal potential residual altitude, if peak value, less than the maximal potential residual altitude, performs step 4, otherwise order classification number adds 1, execution step 3b);

Step 4. merges pseudopotential

4a) by following formula, calculate and obtain histogram gesture partition function:

$F_c\left(k\right)=P_{c-1}^{*}/(1+f_1{(k-x_{c-1})}^4)$

Wherein, F _c(k) expression histogram gesture partition function, subscript c represents c rank histogram gesture partition function, span is c=1,2 ..., L+T, L presentation class number, T represents the pseudopotential number, and k represents number of greyscale levels, and span is 0≤k≤255,

The maximal value that represents image c-1 to be split rank residue potential function, f ₁Represent fuzzy decay factor, x _c-1The gesture extreme point that represents image c-1 to be split rank residue potential function.

4b) will calculate all gesture extreme points corresponding to histogram gesture partition function group that obtain and sort according to ascending order, and according to the clooating sequence of gesture extreme point, corresponding with it histogram gesture partition function be arranged;

4c) judge in histogram gesture partition function group whether have pseudopotential

With in the histogram gesture partition function group after arranging in twos adjacent gesture extreme point corresponding to two gesture partition functions subtract each other, in all differences that obtain when occurring difference less than the value of the fuzzy pseudopotential factor, show in two corresponding with it histogram gesture partition functions and have pseudopotential, perform step 4d), if do not have pseudopotential in histogram gesture partition function group, with this group of functions as image histogram gesture partition function group to be split, the execution step 5;

4d) the pseudopotential in merging histogram gesture partition function group

Extract respectively the extreme value that has two histogram gesture partition functions of pseudopotential in histogram gesture partition function group, relatively two extreme value acquisitions maximal value wherein; , with there being two histogram gesture partition function summations of pseudopotential, obtain one and function; Will be with function by the following formula match, obtain merging the histogram gesture partition function after pseudopotential:

F′(k)＝y/(1+f ₁(k-x) ⁴)

Wherein, F ' (k) represents to merge histogram gesture partition function after pseudopotential, and k represents number of greyscale levels, and span is 0≤k≤255, and y represents to exist the maximal value of two histogram gesture partition functions of pseudopotential, f ₁Represent fuzzy decay factor, x represents the extreme point with function;

4e) all pseudopotentials in histogram gesture partition function group are merged, obtain image histogram gesture partition function group to be split;

Step 5. is obtained cluster centre and the optimal classification number of image to be split

With each gesture extreme point corresponding to image histogram gesture partition function group to be split, as the cluster centre of image to be split, with the optimal classification number of the number of gesture extreme point as image to be split;

Step 6. is with the pixel fuzzy classification of image to be split

Euclidean distance by each pixel and cluster centre in following formula calculating acquisition image to be split:

D _ij＝(I _i-V _j) ²

Wherein, D _ijRepresent the Euclidean distance of i pixel of image to be split to j cluster centre, I _iThe gray-scale value that represents i pixel of image to be split, V _jJ the cluster centre that represents image to be split.

Calculated the degree of membership that obtains each pixel in image to be split by following formula:

$u_{\mathrm{ij}}=1/{\mathrm{\Σ}}_{n=1}^M{(D_{\mathrm{ij}}/D_{\mathrm{in}})}^2$

Wherein, u _ijRepresent that i pixel of image to be split is under the jurisdiction of the degree of membership of j cluster centre, ∑ represents that M represents the optimal classification number of image to be split to expression formula summation, and n represents n class in the optimal classification number of image to be split, D _ijRepresent the Euclidean distance of i pixel of image to be split to j cluster centre, D _inRepresent the Euclidean distance of i pixel of image to be split to n cluster centre.

By the degree of membership of the pixel that calculate to obtain to the classify of image element in image to be split.Method to a certain classify of image element in image to be split is: relatively this pixel, corresponding to the degree of membership of each cluster centre in image to be split, obtains the degree of membership of this pixel maximum, this pixel is classified as the cluster centre that makes its degree of membership maximum.

Step 7. output image segmentation result

To be under the jurisdiction of of a sort pixel mark in image to be split, complete the cutting apart of image, and the output image segmentation result.

Be further described below in conjunction with accompanying drawing 2, accompanying drawing 3,4 pairs of simulated effects of the present invention of accompanying drawing.

1. simulated conditions:

Be to use Matlab7.0a to carry out emulation on Core (TM) 21.86GHZ, internal memory 2G, WINDOWS XP system at CPU.

2. emulation content:

Respectively a width aircraft natural gray image, a width helicopter infrared image and armed people's infrared image are carried out emulation with the fuzzy C-mean algorithm image partition method of the inventive method and prior art, and compared segmentation effect and the splitting speed of two kinds of methods.

3. the simulation experiment result:

3.1. the simulation result of aircraft natural gray image

Respectively the aircraft natural gray image is cut apart with the inventive method and existing fuzzy C-mean algorithm image partition method, the image size is 320 * 400, the optimal classification number that the inventive method self-adaptation obtains the aircraft natural gray image is 2, and the aircraft natural gray image artificially is set respectively in the fuzzy C-mean algorithm image partition method classification number is 2 and 3.The equal independent operating of each experiment 50 times, and consuming time being averaging cut apart in experiment, cutting apart of two kinds of image partition methods of comparison is consuming time as shown in the table:

As can be seen from the above table, no matter whether the artificial classification number that arranges is optimum in the fuzzy C-mean algorithm image partition method, and the inventive method is minimum to image segmentation consuming time.

The segmentation result of aircraft natural gray image as shown in Figure 2, wherein Fig. 2 (a) is the aircraft natural gray image, Fig. 2 (b) is the inventive method segmentation result, Fig. 2 (c) artificially arranges the just segmentation result when the optimal classification number 2 of Images Classification number for the fuzzy C-mean algorithm image partition method, and Fig. 2 (d) is the segmentation result of 3 o'clock for the fuzzy C-mean algorithm image partition method artificially arranges the classification number.As seen from Figure 2, when the fuzzy C-mean algorithm image partition method artificially arranges the Images Classification number just when the optimal classification number 2, the method can obtain the segmentation effect consistent with the inventive method, but when the classification number that image artificially is set is 3, in the method segmentation result, obviously there is more noise spot in the Aircraft Targets edge, and the inventive method energy self-adaptation acquisition image optimal classification number is 2, need not artificial setting, and Aircraft Targets correctly can be cut apart from sky background, Aircraft Targets edge noiseless point, edge is more level and smooth, clear-cut.

3.2. the simulation result of infrared image

Respectively two width infrared images are cut apart with the inventive method and existing fuzzy C-mean algorithm image partition method, wherein, the size of helicopter infrared image is 240 * 320, it is 2 that the inventive method self-adaptation obtains helicopter infrared image optimal classification number, and the helicopter infrared image artificially is set respectively in the fuzzy C-mean algorithm image partition method classification number is 2 and 3; The size of armed people's infrared image is 150 * 200, and it is 3 that the inventive method self-adaptation obtains armed people's infrared image optimal classification number, and armed people's infrared image artificially is set respectively in the fuzzy C-mean algorithm image partition method classification number is 2 and 3; The equal independent operating of each experiment 50 times, and consuming time being averaging cut apart in experiment, cutting apart of two kinds of image partition methods of comparison is consuming time as shown in the table:

As can be seen from the above table, for two width infrared images in experiment, no matter whether the artificial classification number that arranges is optimum in the fuzzy C-mean algorithm image partition method, and the inventive method is minimum to image segmentation consuming time.

The segmentation result of helicopter infrared image as shown in Figure 3, wherein Fig. 3 (a) is the helicopter infrared image, Fig. 3 (b) is the inventive method segmentation result, segmentation result when Fig. 3 (c) artificially arranges helicopter infrared image classification number just for optimal classification number 2 for the fuzzy C-mean algorithm image partition method, Fig. 3 (d) is the segmentation result of 3 o'clock for the fuzzy C-mean algorithm image partition method artificially arranges helicopter infrared image classification number.As seen from Figure 3, when the fuzzy C-mean algorithm image partition method artificially arranges helicopter infrared image classification number just for optimal classification number 2, the method can obtain the segmentation effect consistent with the inventive method, but when the classification number that the helicopter infrared image artificially is set is 3, the method is separated empennage of helicopter part and helicopter fore-end zone with body, be divided into background area, Helicopter Target has been produced obvious mistake minute phenomenon.And the inventive method energy self-adaptation acquisition image optimal classification number is 2, need not artificial setting, and Helicopter Target correctly can be cut apart from background area, in the gained segmentation result, Helicopter Target is more complete, clear-cut, edge effect is more level and smooth, substantially error-free minute phenomenon.

The segmentation result of armed people's infrared image as shown in Figure 4, wherein Fig. 4 (a) is armed people's infrared image, Fig. 4 (b) is the inventive method segmentation result, armed people's infrared image classification number is the segmentation result of 2 o'clock for the fuzzy C-mean algorithm image partition method artificially arranges for segmentation result when image 4 (c) artificially arranges armed people's infrared image classification number just for optimal classification number 3 for the fuzzy C-mean algorithm image partition method, Fig. 4 (d).As seen from Figure 4, when the fuzzy C-mean algorithm image partition method artificially arranges armed people's infrared image classification number just for optimal classification number 3, the method can obtain the segmentation effect consistent with the inventive method, but when the classification number that armed people's infrared image artificially is set is 2, the method is communicated with armed people's target fully with meadow, woods background area, produced obvious mistake minute phenomenon, the gained segmentation result can't be distinguished armed people's objective contour.And the inventive method energy self-adaptation acquisition image optimal classification number is 3, need not artificial setting, and armed people's target can be split clearly from the backgrounds such as river, the woods, in the gained segmentation result, armed people's target is very complete, clear-cut is level and smooth, error-free minute phenomenon.

Can be drawn by above segmentation result, employing is based on the adaptive fuzzy C average image partition method of potential function, the energy self-adaptation obtains the optimal classification number of image to be split, target is correctly cut apart from background area, interior pixels point uniform gray level in target area in the gained segmentation result, the zone consistance keeps better, and the object edge clear-cut is complete, has effectively improved the segmentation effect of image; And the method has increased substantially the splitting speed to image, is conducive to realize the real-time processing to image.

Claims (8)

1. based on the adaptive fuzzy C average image partition method of potential function, concrete steps are as follows:

(1) input image to be split

(2) obtain image histogram potential function to be split and maximal potential residual altitude

2a) adopt the Normalized Grey Level statistic histogram of image to be split, calculate and obtain image histogram potential function to be split;

2b) by calculating the image histogram potential function to be split that obtains, calculate according to the following formula the maximal potential residual altitude:

R＝min{P ₁，P ₂，…，P _n}

Wherein, R represents the maximal potential residual altitude, and span is 0＜R＜1, and min represents the expression formula in { } is minimized, P ₁, P ₂..., P _nThe peak value that represents each crest in image histogram potential function curve to be split, n represent the number of image histogram potential function curve medium wave peak to be split;

(3) obtain image histogram c to be split rank residue potential function

3a) the preliminary classification number of image to be split is set to 2;

3b) calculate fuzzy decay factor and the fuzzy pseudopotential factor;

3c) in image histogram c to be split rank residue potential function group, the initial number of pseudopotential is set to 0;

3d) calculate image histogram c to be split rank residue potential function, and it is poor to calculate in the group of functions that obtains gesture extreme point corresponding to two adjacent in twos rank residue potential functions, when difference is arranged less than the value of the fuzzy pseudopotential factor, just make the pseudopotential number of group of functions add 1 in all differences that obtain;

Whether the peak value that 3e) judges the last single order residue of image histogram to be split potential function is less than the maximal potential residual altitude, if peak value, less than the maximal potential residual altitude, performs step (4), otherwise order classification number adds 1, execution step 3b);

(4) merge pseudopotential

4a) compute histograms gesture partition function;

4b) will calculate all gesture extreme points corresponding to histogram gesture partition function group that obtain and sort according to ascending order, and according to the clooating sequence of gesture extreme point, corresponding with it histogram gesture partition function be arranged;

4c) judge in histogram gesture partition function group whether have pseudopotential

With in the histogram gesture partition function group after arranging in twos adjacent gesture extreme point corresponding to two gesture partition functions subtract each other, in all differences that obtain when occurring difference less than the value of the fuzzy pseudopotential factor, show in two corresponding with it histogram gesture partition functions and have pseudopotential, perform step 4d), if do not have pseudopotential in histogram gesture partition function group, with this group of functions as image histogram gesture partition function group to be split, the execution step (5);

4d) the pseudopotential in merging histogram gesture partition function group

Extract respectively the extreme value that has two histogram gesture partition functions of pseudopotential in histogram gesture partition function group, relatively two extreme value acquisitions maximal value wherein; , with there being two histogram gesture partition function summations of pseudopotential, obtain one and function; Will with function match according to the following formula, obtain merging the histogram gesture partition function after pseudopotential:

F′(k)＝y/(1+f ₁(k-x) ⁴)

Wherein, F ' (k) represents to merge histogram gesture partition function after pseudopotential, and k represents number of greyscale levels, and span is 0≤k≤255, and y represents to exist the maximal value of two histogram gesture partition functions of pseudopotential, f ₁Represent fuzzy decay factor, x represents the extreme point with function;

4e) all pseudopotentials in histogram gesture partition function group are merged, obtain image histogram gesture partition function group to be split;

(5) obtain cluster centre and the optimal classification number of image to be split

With each gesture extreme point corresponding to image histogram gesture partition function group to be split, as the cluster centre of image to be split, with the optimal classification number of the number of gesture extreme point as image to be split;

(6) with the pixel fuzzy classification of image to be split

6a) the Euclidean distance of each pixel and cluster centre in calculating image to be split;

6b) by the degree of membership of calculating pixel in the Euclidean distance calculating image to be split that obtains;

6c) by the degree of membership of the pixel that calculate to obtain to the classify of image element in image to be split;

(7) output image segmentation result

To be under the jurisdiction of of a sort pixel mark in image to be split, complete the cutting apart of image, and the output image segmentation result.

2. the C of the adaptive fuzzy based on potential function average image partition method according to claim 1, is characterized in that step 2a) described image histogram potential function to be split calculates by following formula:

$P\left(k\right)={\mathrm{\Σ}}_{i=0}^{255}H\left(i\right)/(1+\mathrm{\α}{(i-k)}^2)$

Wherein, P (k) represents image histogram potential function to be split, k represents the number of greyscale levels in image histogram potential function to be split, span is 0≤k≤255, ∑ represents expression formula summation, and the Normalized Grey Level statistic histogram of H (i) expression image to be split, i represent the number of greyscale levels in the Normalized Grey Level statistic histogram of image to be split, span is 0≤i≤255, and α represents the Constant control factor.

3. the C of the adaptive fuzzy based on potential function average image partition method according to claim 1, is characterized in that step 3b) described fuzzy decay factor calculates by following formula:

f ₁＝(L-1)/D ²

Wherein, f ₁Represent fuzzy decay factor, L presentation class number, D represent the gradation of image degree of depth to be split, and its value is poor for gray scale maximal value and minimum value in image slices vegetarian refreshments to be split.

4. the C of the adaptive fuzzy based on potential function average image partition method according to claim 1, is characterized in that step 3b) the described fuzzy pseudopotential factor calculates by following formula:

f ₂＝D/3L

Wherein, f ₂Represent the fuzzy pseudopotential factor, D represents the gradation of image degree of depth to be split, and its value is poor for gray scale maximal value and minimum value in image slices vegetarian refreshments to be split, L presentation class number.

5. the C of the adaptive fuzzy based on potential function average image partition method according to claim 1, is characterized in that step 3d) residues potential function in described image histogram c to be split rank calculates by following formula:

$P_c\left(k\right)=P_{c-1}\left(k\right)-P_{c-1}^{*}/(1+f_1{(k-x_{c-1})}^4)$

Wherein, P _c(k) expression image histogram c to be split rank residue potential function, subscript c represents image histogram c to be split rank residue potential function, span is c=1,2 ..., L+T, L presentation class number, T represents pseudopotential number, P _c-1(k) expression image c-1 to be split rank residue potential function,

The maximal value that represents image c-1 to be split rank residue potential function, f ₁Represent fuzzy decay factor, k represents number of greyscale levels, and span is 0≤k≤255, x _c-1The gesture extreme point that represents image c-1 to be split rank residue potential function.

6. the C of the adaptive fuzzy based on potential function average image partition method according to claim 1, is characterized in that step 4a) described histogram gesture partition function calculates by following formula:

$F_c\left(k\right)=P_{c-1}^{*}/(1+f_1{(k-x_{c-1})}^4)$

Wherein, F _c(k) expression histogram gesture partition function, subscript c represents c rank histogram gesture partition function, span is c=1,2 ..., L+T, L presentation class number, T represents the pseudopotential number, and k represents number of greyscale levels, and span is 0≤k≤255,

The maximal value that represents image c-1 to be split rank residue potential function, f ₁Represent fuzzy decay factor, x _c-1The gesture extreme point that represents image c-1 to be split rank residue potential function.

7. the C of the adaptive fuzzy based on potential function average image partition method according to claim 1, is characterized in that step 6a) Euclidean distance of each pixel and cluster centre calculates by following formula in described image to be split:

D _ij＝(I _i-V _j) ²

Wherein, D _ijRepresent the Euclidean distance of i pixel of image to be split to j cluster centre, I _iThe gray-scale value that represents i pixel of image to be split, V _jJ the cluster centre that represents image to be split.

8. the C of the adaptive fuzzy based on potential function average image partition method according to claim 1, is characterized in that step 6b) degree of membership of pixel calculates by following formula in described image to be split:

$u_{\mathrm{ij}}=1/{\mathrm{\Σ}}_{n=1}^M{(D_{\mathrm{ij}}/D_{\mathrm{in}})}^2$

Wherein, u _ijRepresent that i pixel of image to be split is under the jurisdiction of the degree of membership of j cluster centre, ∑ represents that M represents the optimal classification number of image to be split to expression formula summation, and n represents n class in the optimal classification number of image to be split, D _ijRepresent the Euclidean distance of i pixel of image to be split to j cluster centre, D _inRepresent the Euclidean distance of i pixel of image to be split to n cluster centre.

Images