Wrong results while accessing matrix

1 次查看（过去 30 天）
Batuhan Hangun
Batuhan Hangun 2017-7-15
评论： Walter Roberson 2017-7-16
Hello there. I try to implement my own version of SLIC method to work on both an RGB or a Hyperspectral image. After clustering process I need to apply a second step to correct faulty labeling and remove redundant labels. I find the connected components using MATLAB's "bwconncomp" function. Checking connected components sizes and neighbour labels. I want to modify label of some element's label to their nearest neighbour cluster center's label. To achieve this I need to find an average vector [L a b x y] of given connected component. Labxy is a vector that I created for this purpose. But the values those are stored in Labxy vector are not correct. What am I doing wrong?
clc,clear, close all;
Image = imread('anyrgbimage.jpg');
Superpixels = 512;
tic
[LabelMatrix,ClusterCenters] = slic(Image, Superpixels,10);
toc
ImageLab = rgb2lab(Image);
[Height, Width] = size(LabelMatrix);
SuperPixelSize = (Height * Width)/ Superpixels;
MinimumSegmentSize = round(SuperPixelSize/4);
LabelMatrix2 = LabelMatrix;
for label = 1:size(ClusterCenters,1)
% fprintf('Calculations for Label %d\n\n', label');
ConnectedComponentsL = bwconncomp(LabelMatrix == label);
numPixels = cellfun(@numel,ConnectedComponentsL.PixelIdxList);
[BiggestSegment,idx] = max(numPixels);
for i = 1:length(numPixels)
if(BiggestSegment > MinimumSegmentSize)
if(i ~= idx )
% fprintf( 'numPixels(%d) = %d\n\n', i,numPixels(i));
Segment = ConnectedComponentsL.PixelIdxList{i};
S = [Height , Width];
AdjacentLabelCount = zeros(1,size(ClusterCenters,1));
Labxy = uint8(zeros(length(Segment),size(ClusterCenters,2)));
for j = 1:length(Segment)
[X, Y] = ind2sub(S, Segment(j));
Labxy(j,:) = [(squeeze(ImageLab(X,Y,:)))' X Y];
fprintf('[%d %d %d %d %d]\n\n', Labxy(1),Labxy(2),Labxy(3), X, Y);
fprintf('%d) Segment Index = %d, Subscript = [%d %d], Size of Segment = %d elements \n\n', j, Segment(j), X, Y, numPixels(i));
NLoc = [X, Y] + [-1 , 0];
% fprintf('NLoc = [%d %d] ', NLoc(1), NLoc(2));
NELoc = [X, Y] + [-1 , 1];
% fprintf('NELoc = [%d %d] ', NELoc(1), NELoc(2));
NWLoc = [X, Y] + [-1 , -1];
% fprintf('NWLoc = [%d %d] ', NWLoc(1), NWLoc(2));
SLoc = [X, Y] + [ 1 , 0];
% fprintf('SLoc = [%d %d] ', SLoc(1), SLoc(2));
SELoc = [X, Y] + [1 , 1];
% fprintf('SELoc = [%d %d] ', SELoc(1), SELoc(2));
SWLoc = [X, Y] + [1 , -1];
% fprintf('SWLoc = [%d %d] ', SWLoc(1), SWLoc(2));
WLoc = [X, Y] + [0 ,-1];
% fprintf('WLoc = [%d %d] ', WLoc(1), WLoc(2));
ELoc = [X, Y] + [0 , 1];
% fprintf('ELoc = [%d %d] \n\n', ELoc(1), ELoc(2));
[AdjacentLabelCount, NorthLabel] = checknorth(NLoc, Height, Width, label, LabelMatrix, AdjacentLabelCount);
[AdjacentLabelCount, NorthWestLabel] = checknorthwest(NWLoc, Height, Width, label, LabelMatrix, AdjacentLabelCount);
[AdjacentLabelCount, NorthEastLabel] = checknortheast(NELoc, Height, Width, label, LabelMatrix, AdjacentLabelCount);
[AdjacentLabelCount, SouthLabel] = checksouth(SLoc, Height, Width, label, LabelMatrix, AdjacentLabelCount);
[AdjacentLabelCount, SouthWestLabel] = checksouthwest(SWLoc, Height, Width, label, LabelMatrix, AdjacentLabelCount);
[AdjacentLabelCount, SouthEastLabel] = checksoutheast(SELoc, Height, Width, label, LabelMatrix, AdjacentLabelCount);
[AdjacentLabelCount, EastLabel] = checkeast(ELoc, Height, Width, label, LabelMatrix, AdjacentLabelCount);
[AdjacentLabelCount, WestLabel] = checkwest(WLoc, Height, Width, label, LabelMatrix, AdjacentLabelCount);
%MaxMat = find(AdjacentLabelCount == max(AdjacentLabelCount));
end
MaxMat = find(AdjacentLabelCount == max(AdjacentLabelCount));
if(length(MaxMat) == 1)
% fprintf('Biggest adjacent of Label %d is Label %d\n\n', label, MaxMat);
LabelMatrix2(Segment) = MaxMat;
else
% fprintf('There are more than one adjacent labels at same quantity\n\n');
for t = 1 : length(MaxMat)
% fprintf('%d. Biggest adjacent of Label %d is Label %d\n\n', t, label, MaxMat(t))
% LabelMatrix2(Segment) = MaxMat;
end
end
end
else
fprintf('Segment size of Label %d is equal to %d and it is smaller than Minimum Segment Size which is %d\n\n', ...
label, numPixels(i) , MinimumSegmentSize);
end
end
end

请先登录，再回答此问题。

回答（1 个）

Walter Roberson
Walter Roberson 2017-7-15
You have
[X, Y] = ind2sub(S, Segment(j));
Labxy(j,:) = [(squeeze(ImageLab(X,Y,:)))' X Y];
You have made the fairly common mistake of thinking that rows correspond to X and that columns correspond to Y. Instead, rows correspond to Y and columns correspond to X. So,
[Y, X] = ind2sub(S, Segment(j));
Labxy(j,:) = [(squeeze(ImageLab(Y,X,:)))' X Y]; %you need to rethink here whether you want X Y or Y X at the end
  2 个评论
Batuhan Hangun
Batuhan Hangun 2017-7-16
Actually my indexing was right I guess because I did not get any error about matrix dimensions. I think source of wrong results was uint8( ) at
Labxy=uint8(zeros(length(Segment),size(ClusterCenters,2)));
I removed that conversion and everything seems fine. Now I try to access ImageLab(X,Y,:) with [X Y] given by ind2sub(S, Segment(j)); and it looks like that it does fine. I think while displaying it(imagesc, image, imshow), MATLAB converts ImageLab(X,Y,:) to ImageLab(Y,X,:) right? Since I do not display anything
[X, Y] = ind2sub(S, Segment(j));
Labxy(j,:) = [(squeeze(ImageLab(X,Y,:)))' X Y];
this usage is okay. Am I right?
Walter Roberson
Walter Roberson 2017-7-16
MATLAB always uses rows for Y and columns for X for image display. The only question for MATLAB is whether location (1,1) should represent the top left of the image or the bottom left of the image.

请先登录，再进行评论。

类别

在 Help Center 和 File Exchange 中查找有关 Linear Algebra 的更多信息

标签

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by