remove non-circle objects from image

It looks that I tried everything (eccentricity, bwareaopen and so on)but nothing works. Maybe someone could suggest a way to do it?

I am dealing with this image (two styles of representation):

http://postimage.org/image/ojief2spl/

http://postimage.org/image/t6oggug2h/

shapes=[stats.Eccentricity];
removeA2=bwareaopen(shapes<0.20,100);

or

bw = bwareaopen(bw, 50);

and others. I just can't find it in my code. I think =edge(); didn't worked well enough.

[Sean de Wolski]

Well first off, you did not try _everything_. What have you tried? Show us, in code, and give us a link to an image that you have so we can see what you're working with.

Also, be sure to check out ImageAnalyst's "blobsdemo" on the fex.

[Sean de Wolski]

What about 'extent' or perimeter to area ratios? These are all easy options in regionprops.

[LG9]

I tried to read about it but I couldn't do it by myself

[LG9]

I found those ratios but I don't know how to use them to remove objects

[Sean de Wolski]

Well how are you going to learn to solve problems without working through them yourself? Take a look at the 'extent' of a few of the blobs, see if you notice a pattern. Or look at the area to perimeter ratio, and once again look for a pattern.

[LG9]

http://postimage.org/image/yruinlal7/

[Sean de Wolski]

Good job! There's a pretty obvious pattern there, threshold those values and keep the high ones.

[LG9]

so I am trying to set threshold value and then use bwareaopen(). Where is my mistake?

threshold = 0.8;

F = bwareaopen(fillA,threshold);

imshow(F)

Answer 1

[Sven]

% Get the image as a b&w indexed (non-rgb) image
I = imread('project.jpg');
BW = rgb2ind(I, 2);

% Calculate its connected regions
L = labelmatrix(BW); % Not using bwconncomps() for older version users
stats = regionprops(CC,'Extent','Area');

% Find the ones that are like a circle
minExtent = 0.75;
keepMask = [stats.Extent]>minExtent;

% Extract the image of circles only and display
BWcircles = ismember(L, find(keepMask));
BWnonCircles = BW & ~BWcircles;

% Show the circles
figure, imshow(BWcircles)

% Show the things we removed
figure, imshow(BWnonCircles)

% Show them together in different colours
RGB = double(cat(3, zeros(size(BW)), BWcircles, BWnonCircles));
figure, imshow(RGB)

[LG9]

Could function below be available only on the newest version of program? It looks that it is not recognised by a Matlab which is in my university

CC = bwconncomp(BW);

[Sean de Wolski]

What version are you using? Type >> ver at the command line.

[Sven]

You may have a version before bwconncomp()

It can equivalently be:

BW = rgb2ind(I, 2);

L = bwlabel(BW);

stats = regionprops(L, ...)

[LG9]

thanks, now it works on the version I have but this code identifies circle objects. I think I was able to do this before. I just don't know how to remove other objects who has coefficient <0.7 for example

[Sven]

LG9, I chose to show you all circles (stored in BWcircles) and all non-circles (stored in BWnonCircles), just in different colours. If you want to "remove non-circles", then all you have to do is look at the BWcircles image by itself:

figure, imshow(BWcircles)

[Sven]

LG9, if your question has been answered, hit "accept" to remove it from the "unanswered" list.