# Estimation of the "pixelization" error introduced when using the digital image processing toolbox

2 views (last 30 days)
Miguel on 22 May 2012
Hello dear Matlab users,
I am using the digital image processing toolbox in order to obtain the equivalent diameter of objects located in an image. The objects will be roundish, most of the time; so using the regionprops function works fine for me. However, I know that Matlab estimates the equivalent diameter based on the area of each one of the objects contained in the image. My question is about the estimation of the error incurred by Matlab in rounding the empty/filled pixels. I do not know how to assess this situation due to the fact that I do not know the exact diameter of the objects, so I have no point of comparison to determine the deviation from the exact value and the Matlab value.
Also, the original objects have no jagged edges but after processing the images with Matlab, most of the objects will have jagged edges.
I hope that my situation is explained in a clear manner.
Thanks for any kind of help, info or feedback, all of them will be gladly received.
Regards,
-- MB
Walter Roberson on 22 May 2012

Geoff on 22 May 2012
This is just a stab in the dark, but I gather that after processing you get an image with logical (0 or 1) pixels, and you obtain a centre point and diameter for your round objects.
Could you not simply count the number of pixels inside your shape (discrete area), and then compare that to the area of the circle that you have detected?
Miguel on 22 May 2012
@ Walter: Exactly! If I knew the diameter beforehand it would be a simple task. But the question derives from the fact that I am using Matlab to estimate the diameter of the objects in the image; and I was wondering how accurate is this estimation.
Geoff on 22 May 2012
Oh yep, my mistake... In that case I would probably go back to the original image and sum up the pixels at the edge of the processed object. First I'd extract the 'edge' by taking all pixels where in any 2x2 square there are both 0 and 1 values. Then I'd look at the intensity values for those pixels in the squares matched with their on/off processed counterpart.... You're wanting to measure how much aliasing was introduced by the processing. Might not be practical depending on background etc....