分割图像中特定片段的邻居

Question

考虑分割算法的输出是一个与输入图像大小相同的矩阵。对于检测到的每个段，矩阵的特定部分都使用特定的数字/索引进行索引，如下所示。

我想检索特定段的邻居。 Here 有描述取每个像素不同大小的邻居。例如，在 3x3、5x5、7x7 等中。我们可以使用上面链接中的时间高效方式对不同大小（大约）的邻域中的特定段执行类似的过程吗？

最好的问候，

透特

PS：非常感谢任何问题。

Answer 1

如果要查找特定区域的邻居的索引，可以执行以下操作：

%# input: lblImg - image where pix(x,y) is the index of the segment
targetSegment = 4;
%# grow the segment by 2 pixels, since I assume the boundary in between is 1 pixel wide
msk = imdilate(lblImg == targetSegment,strel('disk',2));
msk(lblImg == targetSegment | lblImg == 0) = false; %# remove original cell, and boundary

%# retrieve the list of neighbors
listOfNeighbors = unique(lblImg(msk));

当然，您可以对所有段并行执行此操作，并创建一个邻接矩阵。 Dilation 是一个局部极大值操作，所以它会使不同索引的标签重叠。

 dilImg = imdilate(lblImg, strel('disk',2));
 msk = dilImg ~= lblImg & lblImg > 0; %# assume indices are all positive

 %# msk contains the indices of pixels where dilation
 %# has created an overlap between segments.

 rowColIdx = unique( [dilImg(msk), lblImg(msk)], 'rows');

 %# create adjacency matrix. Due to the nature of imdilate, this will fill in 
 %# only the values below the diagonal.
 %# adjacencyMatrix = adjacencyMatrix + adjacencyMatrix.' would fix that.
 nLabels = max(lblImg(:));
 adjacencyMatrix = sparse(rowColIdx(:,1),rowColIdx(:,2),ones(size(rowColIdx,1),1),...
   nLabels, nLabels);