How to arrange the shape# (Blob#) in proper order (ascending/descending)?

Hello, I am using matlab to recognize the all square object. I need centriods of each square. I am using the following:

    [labeledImage, numberOfObjects] = bwlabel(binaryImage);
    s = regionprops(labeledImage,'centroid');

to get the centriods. But these shape numbers are so random and I am struggling to arrange it in any proper way. I have to dot it manually which is quite time consuming. Any expert comments will be highly appreciated.

<<

>>

[Saqib Sharif]

My complete Code:

function mesh
ImageFile=('GIST_S16_03.tif');
RGB = imread(ImageFile);
GRAY = rgb2gray(RGB);
threshold = adaptthresh(GRAY,0.613);
BW = imbinarize(GRAY, threshold);
BW = ~ BW;
binaryImage = imfill(BW,'holes');
binaryImage = bwareaopen(binaryImage, 800);
[labeledImage, numberOfObjects] = bwlabel(binaryImage);
filledImage = imfill(binaryImage, 'holes');
boundaries = bwboundaries(filledImage);
figure
imshow(labeledImage); 
title('Binary Image with centriods');
hold on
s = regionprops(labeledImage,'centroid');
centroids = cat(1, s.Centroid);
x=centroids(:,1);
y=centroids(:,2);
plot(x,y, 'r*');
for blobNumber = 1 : numberOfObjects
    thisBoundary = boundaries{blobNumber};
    plot(thisBoundary(:,2), thisBoundary(:,1), 'r', 'LineWidth', 2);
      overlayMessage = sprintf('%d',blobNumber);
      text(centroids(blobNumber,1), centroids(blobNumber,2), ...
    overlayMessage, 'Color', 'g','FontSize',14);
end

Answer 1

[Jim Joy]

Hi Saqib,

The labeled image that you have shown is consistent with the conventions that "bwlabel" uses to label the image. The labeling occurs like this because "bwlabel" searches down each column, and labels each region based on when it encounters a pixel first. That is, regions with pixels further leftward than others will be counted first. You can read more about this in the blog post linked below:

https://blogs.mathworks.com/steve/2008/04/14/relabeling-a-label-matrix/

To relabel the image regions in the manner that you would like, I would recommend looping over the columnar structure of your labeled image, and relabeling the regions based on the y-positions of the centroid. For example, the first 'column' would be transformed from [8, 9, 6, 5, 7, 3, 4, 2] to [2, 3, 4, 5, 6, 7, 8, 9]. Please note that looping in this fashion will require some level of filtering to divide the image into columns.

I hope this helps.

Jim

[Walter Roberson]

      s = regionprops(labeledImage,'centroid');
      cents = vertcat(s.Centroid);
      [~, sortidx] = sortrows(cents);
      s = s(sortidx);

Though as Jim points out you might want to do some smoothing on the x coordinates, or some quantization.