Why regionprops some values looks like bugs?

I have Tried Using regionprops with fields of Area, Centroid and BoundingBox. I was desappointed firstly when comparing results of my implementation of Centroid vs the ready Centroid field in regionProps when I find that they are sometimes mostly equally, but sometimes the regionprops centroids give very wrong results.

As for prove here is the values I get from boundingbox

value =

     53.5000   36.5000  170.0000  635.0000

value =

     41.5000   10.5000  152.0000  671.0000

value =

     27.5000   51.5000  217.0000  622.0000

value =

     49.5000    7.5000  177.0000  686.0000

value =

     35.5000  631.5000    1.0000    1.0000

value =

     49.5000   23.5000  180.0000  656.0000

value =

     33.5000  471.5000    1.0000    1.0000

value =

     41.5000   26.5000  207.0000  674.0000

value =

     43.5000  358.5000    1.0000    1.0000

value =

     46.5000   59.5000  204.0000  616.0000

value =

     33.5000   36.5000  214.0000  632.0000

value =

     31.5000  287.5000    1.0000    1.0000

value =

     28.5000   29.5000  182.0000  646.0000

value =

     56.5000    1.5000  187.0000  699.0000

Why are those 1's in the result??? it's not right at all.

Answer 1

[Image Analyst]

It's not a bug. Those particular blobs are only 1 pixel big, so the width and height are only 1 pixel. The width and height are measured in terms of whole pixels, while the bounding box goes from a half pixel to the left of the left-most pixel center to a half pixel to the right of the right-most pixel center.

A = [0 1 0 1 1 0]
measurements = regionprops(logical(A), A, 'Area', 'BoundingBox');
for k = 1 : length(measurements)
  fprintf('For blob #%d\n  the area = %.1f, \n  and the bounding box = [%f, %f, %f, %f]\n',...
  k, measurements(k).Area, measurements(k).BoundingBox); 
end

and see

For blob #1
  the area = 1.0, 
  and the bounding box = [1.500000, 0.500000, 1.000000, 1.000000]
For blob #2
  the area = 2.0, 
  and the bounding box = [3.500000, 0.500000, 2.000000, 1.000000]

[Eslam Hamed]

I know about 'BoundingBox' that regionprops

But I'm still confused, could you please simple it a little?

[Image Analyst]

It returns a structure array. Each structure has members/fields that are the measurements you asked for. You must know all this already if you're already using regionprops. Do you have anything more specific you can ask?

[Eslam Hamed]

No, Thanks a lot it I got it you were right ^_^

[Eslam Hamed]

Thank you again, my bad was my image processing steps, I just used imclose to remove the noise that was wrong from me, thank you a lot again :)

[Image Analyst]

You can use bwareaopen(binaryImage, 1) or bwmorph(binaryImage, 'clean') to remove single pixels.