MATLAB中图像之间的Mutual information的计算

最新推荐文章于 2022-03-16 22:48:32 发布

脉望虫

最新推荐文章于 2022-03-16 22:48:32 发布

阅读量1.9k

点赞数 1

分类专栏：图像处理文章标签： matlab opencv python

本文链接：https://blog.csdn.net/qq_36763031/article/details/108944800

版权

图像处理专栏收录该内容

11 篇文章

订阅专栏

本文介绍了如何在图像处理中利用Mutual Information（MI）计算公式来量化两张图像之间的相似性，通过量化、直方图和联合分布的计算，详细展示了Matlab代码示例。重点讲解了联合直方图、边际熵和联合熵的计算，以及MI的实际应用和计算过程。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

Mutual information

图像中Mutual information的计算公式一般如下：
在这里插入图片描述
这之中涉及到两个方面的内容，一个是联合的直方图，一个是各自的直方图。下面是参考代码：

function res = computeMI(img1,img2,bins)
%Compute mutual information
%Quantify images
img1 = im2double(uint16(img1));
img2 = im2double(uint16(img2));
img1 = floor(img1*bins);%Quantify
img2 = floor(img2*bins);

%参考：
%https://www.mathworks.com/matlabcentral/fileexchange/6978-image-registration-2d-using-mutual-information-optimization-toolbox-needed
[rows,cols] = size(img1);

h = hist2d ([img1(:),img2(:)], 0:1:100, 0:1:100);

[r,c] = size(h);
b= h./(r*c); % normalized joint histogram
y_marg=sum(b); %sum of the rows of normalized joint histogram
x_marg=sum(b');%sum of columns of normalized joint histogran

Hy=0;
for i=1:c    %  col
    if( y_marg(i)==0 )
        %do nothing
    else
        Hy = Hy + -(y_marg(i)*(log2(y_marg(i)))); %marginal entropy for image 1
    end
end

Hx=0;
for i=1:r    %rows
    if( x_marg(i)==0 )
        %do nothing
    else
        Hx = Hx + -(x_marg(i)*(log2(x_marg(i)))); %marginal entropy for image 2
    end   
end
h_xy = -sum(sum(b.*(log2(b+(b==0))))); % joint entropy

res=-(Hx+Hy-h_xy);% Mutual information
%x
end

% https://www.mathworks.com/matlabcentral/fileexchange/1487-2d-histogram-matrix
function mHist = hist2d (mX, vYEdge, vXEdge)
nCol = size(mX, 2);
if nCol < 2
    error ('mX has less than two columns')
end
nRow = length (vYEdge)-1;
nCol = length (vXEdge)-1;
vRow = mX(:,1);
vCol = mX(:,2);
mHist = zeros(nRow,nCol);
for iRow = 1:nRow
    rRowLB = vYEdge(iRow);
    rRowUB = vYEdge(iRow+1);
    
    vColFound = vCol((vRow > rRowLB) & (vRow <= rRowUB));
    
    if (~isempty(vColFound))
        
        
        vFound = histc (vColFound, vXEdge);
        
        nFound = (length(vFound)-1);
        
        if (nFound ~= nCol)
            disp([nFound nCol])
            error ('hist2d error: Size Error')
        end
        
        [nRowFound, nColFound] = size (vFound);
        
        nRowFound = nRowFound - 1;
        nColFound = nColFound - 1;
        
        if nRowFound == nCol
            mHist(iRow, :)= vFound(1:nFound)';
        elseif nColFound == nCol
            mHist(iRow, :)= vFound(1:nFound);
        else
            error ('hist2d error: Size Error')
        end
    end
    
end
end

链接中也给出了引用地址。