Discarding certain centroids found using regionprops

Hi, I work with time sequence images. I process and threshold these images to get a black image with some scattered white spots. I want to track these white spots as a function of time and I want to use their centroids to do so.

I am using the functions 'bwlabel' and 'regionprops' to find the centroid and area of the white spots. The problem I am facing is that regionprops also computes the centroids and areas of certain small white specks that are in the black region of my image. I see that all of these points have a unit area. These specks may be just noise.

Hence I was wondering if someone could tell me what function I could use to remove these unwanted points. I know their areas and their position in the regionprops matrix.

Thanks

NS

If 'I' is my processed double image, my code is as follows:

L=bwlabel(I);
s=regionprops(L,'Centroid','Area');

Answer 1

[Sean de Wolski]

What do you mean it computes the centroid parts in a black area? There's no way that it does that.

It's possible the centroid of a blob isn't in the blob. But it's definitely not calculating centroids for the background. If you think of a blob in the shape of "U", the centroid will not lie in on the U.

If you want to keep only centroids that occur in the area of a blob, that's easily possible.

EDIT Per comment/clarification:

I would just use bwareaopen on your logical image to remove small blobs.

I = bwareaopen(I,10); %all objects with fewer than 10px gone.

[NS]

You are correct. Those unwanted centroids are for tiny white specks that are still there in my thresholded image. I think these specks are just noise. So is there any way I could discard those centroids?

[NS]

It worked. bwarea removed the unwanted blobs. Thanks Sean. :)

Answer 2

[Aaron Greenbaum]

Hello NS, I'm actually doing something that sounds pretty similar. It sounds like these unwanted centroids probably have smaller area than your desired centroids. I remove all of the points that are below a certain minimum area using a logical vector.

smoothimg=filter2((ones(2)/2.^2),inimg); %Smooth image
threshimg=smoothimg>thres; %apply threshold
stats = regionprops(threshimg,smoothimg, 'Area', 'WeightedCentroid');
rel_peaks_vec=[stats.Area]>=minarea; %makes logical vector for points greater than minarea as true
cents= [stats(rel_peaks_vec).WeightedCentroid]'; %appostrophe causes a transpose of the matrix

Maybe this can give you some ideas. Just realized this was 5 years old. My bad haha.