[Image segmentation] Image segmentation based on matlab random walk algorithm [including Matlab source code 084]

1. Introduction

About global optimization Solving
global optimization is a very complex problem, and there is no universal method to solve the global optimal value for any complex function. The previous article explained a method for solving local minimums-gradient descent method. This method is practical for situations where the accuracy of the solution is not high, and the local minimum can be used to approximate the global minimum. But when it is required to accurately solve the global minimum, the gradient descent method is not applicable, and other methods are needed to solve it. Common methods to solve the global optimization include Lagrangian method, linear programming method, and some artificial intelligence algorithms such as genetic algorithm, particle swarm algorithm, simulated annealing algorithm, etc. (see my previous blog). But today I’m going to talk about a method that is simple to operate but not easy to fall into a local minimum: the random walk algorithm.
Operation steps
of random walk algorithm Let f(x)f(x) be a multivariate function with nn variables, x=(x1,x2,...,xn)x=(x1,x2,...,xn) is nn dimensions vector.
Given the initial iteration point xx, the first walking step λλ, the control accuracy ϵϵ (ϵϵ is a very small positive number, used to control the end of the algorithm).
Given the number of iteration control NN, kk is the current iteration number, set k=1k=1.
When k<Nk<N, randomly generate a nn-dimensional vector between (−1,1)(−1,1) u=(u1,u2,...,un),(−1<ui<1,i =1,2,...,n)u=(u1,u2,...,un),(−1<ui<1,i=1,2,...,n), and standardize it to get u′=u∑ ni=1u2i√u′=u∑i=1nui2. Let x1=x+λu′x1=x+λu′ to complete the first step of walking.
Calculate the value of the function, if f(x1)<f(x)f(x1)<f(x), that is, find a better point than the initial value, then reset kk to 1, change x1x1 to xx, and return Step 2; otherwise k=k+1k=k+1, go back to step 3.
If no better value can be found for consecutive NN times, it is considered that the optimal solution is in the NN sphere with the current optimal solution as the center and the current step size as the radius (if it is three-dimensional, it just happens to be in space Sphere). At this point, if λ<ϵλ<ϵ, the algorithm ends; otherwise, let λ=λ2λ=λ2, go back to step 1, and start a new round of travel.

Second, the source code

clear ;
close all;
 
addpath 'algorithms'
out = ['results\'];
if ~exist(out)
    mkdir(out);
end
 
%% parameters            
only_name= '41004';
img_name = ['./imgs/' only_name '.jpg'];
ref_name = ['./scribbles/' only_name '.bmp'];
 
 
nei = 1;            % 0: 4-neighbors, 1: 8-neighbors
c = 0.0004;%1e-3;%         % restarting probability of RWR
sigma_c = 60;       % color variance
scale = 1.0;        % image resize
lambda = 2e-10;     % parameter for unitary
isKeepConnect = 0;  % 1: only consinder the connected regions with seeds;  0: otherwise.
reset(RandStream.getGlobalStream); % fix the random seed for initalization of GMM
 
saveProb = 0; % 1: save the probability image
%% main routine
img = imread(img_name); img = imresize(img,scale);
[K, labels, idx] = seed_generation(ref_name,scale);
 
%% RWR with prior
% run('vlfeat-0.9.13/toolbox/vl_setup');
st=clock;
[posteriors label_img] = do_RWR_prior(img,idx,labels,c,lambda,nei,sigma_c,isKeepConnect);
fprintf('subRW took %.2f second\n',etime(clock,st));
 
% display
[imgMasks,segOutline,imgMarkup]=segoutput(im2double(img),label_img); %clear imgMasks segOutline;
 
outPath = [out,'ours\'];
if ~exist(outPath)
    mkdir(outPath);
end
 
figure; clf;set(gcf,'Position',[100,500,size(img,2)*(K+1),size(img,1)]);
for k=1:K 
    prob_img = sc(posteriors(:,:,k),'prob_jet');
    if saveProb == 1
        imwrite(prob_img,[outPath,only_name,'_prob',num2str(k),'.png']);
    end
    subplot(1,K+1,k); imshow(prob_img); clear prob_img;
end;
subplot(1,K+1,K+1); imshow(imgMarkup);
figure,imshow((K-imgMasks)/(K-1));
 
imwrite(imgMarkup,[outPath,only_name,'_bound.png']);
imwrite((K-imgMasks)/(K-1),[outPath,only_name,'_binary.png']);

Three, running results

Insert picture description here

Four, remarks

[Image segmentation] Image segmentation based on matlab random walk algorithm [including Matlab source code 084]

1. Introduction

Second, the source code

Three, running results

Four, remarks

Guess you like