基于matlab对比度和结构提取的多模态解剖图像融合实现

程序员文章站 2022-04-29 10:49:42

目录一、图像融合简介二、部分源代码三、运行结果四、matlab版本一、图像融合简介应用多模态图像的配准与融合技术,可以把不同状态的医学图像有机地结合起来,为临床诊断和治疗提供更丰富的信息。介绍了多模态...

一、图像融合简介

应用多模态图像的配准与融合技术,可以把不同状态的医学图像有机地结合起来,为临床诊断和治疗提供更丰富的信息。介绍了多模态医学图像配准与融合的概念、方法及意义。最后简单介绍了小波变换分析方法。

二、部分源代码

clear; close all; clc; warning off
%% a novel multi-modality anatomical image fusionmethod based on contrast and structure extraction
% f = fuseimage(i,scale)

%inputs:
%i - a mulyi-modal anatomical image sequence

%scale - scale factor of dense sift, the default value is 16

%% load images from the folder that contain multi-modal image to be fused
%i=load_images('./dataset\ct-mri\pair 1');
i=load_images('./dataset\mr-t1-mr-t2\pair 1');
%i=load_images('./dataset\mr-gad-mr-t1\pair 1');
% show source input images 
figure;
no_of_images = size(i,4);
for i = 1:no_of_images
    subplot(2,1,i); imshow(i(:,:,:,i));
end
suptitle('source images');


%%
f=fuseimage(i,16);
%% output: f - the fused image

f=rgb2gray(f);
figure;
imshow(f);
function [ f ] = fuseimage(i,scale)


addpath('pyramid_decomposition');
addpath('guided_filter');
addpath('dense_sift');

tic
%%
[h, w, c, n]=size(i);
imgs=im2double(i);
ia=zeros(h,w,c,n);
for i=1:n
ia(:,:,:,i)=enhnc(imgs(:,:,:,i));

end
%%
imgs_gray=zeros(h,w,n);
for i=1:n
    imgs_gray(:,:,i)=rgb2gray(ia(:,:,:,i));
end
%
% %dense sift calculation
dsifts=zeros(h,w,32,n, 'single');
for i=1:n
    img=imgs_gray(:,:,i);
    ext_img=img_extend(img,scale/2-1);
    [dsifts(:,:,:,i)] = densesift(ext_img, scale, 1);
    
end
%%
%local contrast
contrast_map=zeros(h,w,n);
for i=1:n
    contrast_map(:,:,i)=sum(dsifts(:,:,:,i),3);

end


%winner-take-all weighted average strategy for local contrast

[x, labels]=max(contrast_map,[],3);
clear x;
for i=1:n
    mono=zeros(h,w);
    mono(labels==i)=1;
    contrast_map(:,:,i)=mono;

end



%% structure 
h = [1 -1];
structure_map=zeros(h,w,n);

for i=1:n
structure_map(:,:,i) = abs(conv2(imgs_gray(:,:,i),h,'same')) + abs(conv2(imgs_gray(:,:,i),h','same')); %eq 13

   
end


%winner-take-all weighted average strategy for structure

[a, label]=max(structure_map,[],3);
clear x;
for i=1:n
    monoo=zeros(h,w);
    monoo(label==i)=1;
    structure_map(:,:,i)=monoo;
     
end

%%
weight_map=structure_map.*contrast_map;




%weight map refinement using guided filter
for i=1:n
    
    weight_map(:,:,i) = fastgf(weight_map(:,:,i),12,0.25,2.5);
 
end



% normalizing weight maps
%
weight_map = weight_map + 10^-25; %avoids division by zero
weight_map = weight_map./repmat(sum(weight_map,3),[1 1 n]);

%% pyramid decomposition

% create empty pyramid
pyr = gaussian_pyramid(zeros(h,w,3));
nlev = length(pyr);

% multiresolution blending
for i = 1:n
    % construct pyramid from each input image
   
    % blend
    for b = 1:nlev
        w = repmat(pyrw{b},[1 1 3]);
        
        pyr{b} = pyr{b} + w .*pyri{b};
    end
    
end

% reconstruct
f = reconstruct_laplacian_pyramid(pyr);

toc

end