How to use the 'imfilter' function to pick up blobs that are circular/linear

I'm working on an image analysis project based around blob detection. This question has got to do with filter adjustment. The script I'm using picks up blobs that are more of a circular shape. However some images I'm now working with have blobs shaped more linear (Or even blobs that generally have an irregular shape). Below is the code I am currently using and it is a low pass filter. I've been messing about with it to try and pick up the irregular shapes:

PSF = fspecial('gaussian',7,7);                                        
 %'fspecial' creates predefined 2D filter, in this case a gaussian low pass filter 

Blurred = imfilter(double(image_adapthisteq),PSF,'conv');               
%imfilter 'conv' performs multidimensional filtering using convolution. Data type = 'double'

figure, imshow(uint8(Blurred)), title('Blurred')

I'm thinking the problem could have got to do with the gaussian LPF, however what substitute could you use if you still want to pick up circular blobs at the same time?

Or in the second line of coding 'conv' but I've tried replacing it and can't get any where! Any ideas where an adjustment can be made?

Answer 1

[Image Analyst]

The way I do it, and I think is more common, is to compute the perimeter squared to area ratio:

labeledImage = bwlabel(binaryImage);
measurements = regionprops(labeledImage, 'Area', 'Perimeter');
allAreas = [measurements.Area];
allPerimeters = [measurements.Perimeter];
circularities = allPerimeters .^ 2 ./ (4 * pi * allAreas);
roundObjectsIndexes = find(circularities < 3); % or whatever value you want.
% Extract only those blobs that meet our criteria, and
% eliminate those blobs that don't meet our criteria.
% Note how we use ismember() to do this.
keeperBlobsImage = ismember(labeledImage, keeperIndexes); % A labeled image
% If you want a binary image also, threshold the labeled image
binaryImageOfRoundObjects = keeperBlobsImage > 0;

[Cormac]

With respect to area and perimeter, there is no set dimensions to the blobs. One could say it is impossible to have a blob of the same measurement to appear twice in different images. They are usually all unique. Would the above code still be viable for a scenario as such.

Thanks for getting back to me also

[Troy Steiner]

If I understand this correctly, this should remove circular blobs from a photo?

Should this be a working script?

[Image Analyst]

Cormac. I don't know why you say that. Obviously it's possible to have blobs in different pictures have the same area - they just have to have the same number of pixels. If it's the same scene/objects and you have good control over lighting and positioning, it's not unusual to have areas that are very close, if not exact. So it is possible. And of course my code would still work. It just looks at blobs in one picture - each picture is an independent situation. It doesn't care what's in any other picture. You can keep or discard round blobs in one or more pictures.

Tony, the script is a partial script, not a full demo. It's assuming that you did image segmentation first to produce a binary image. Once you have that, then you can follow with the code in my script.