Segmentate a circular object with hough transform

I am working on a classifier of yoyos with different types of images and different types of backgrounds. For instance, I have this image

so my idea was to detect edges and then apply the hough transform for detecting circles implemented by matlab imfindcircles but it just detects other small circles. This is my code:

I = imread('249_0084.jpg'); I = rgb2gray(I); I = im2double(I);

h = fspecial('gaussian',[10 10],5); I2 = abs(imfilter(I,h,'replicate'));

h = fspecial('unsharp'); I2 = abs(imfilter(I,h,'replicate'));

I2 = edge(I2,'zerocross'); I2 = bwmorph(I2,'diag'); I2 = bwmorph(I2,'close'); r_range = [15 floor(min(size(I2))/2)]; [centers radii] = imfindcirles(I2, r_range);

Any help or idea of how could I proceed?

[Image Analyst]

Are all your yoyos going to be uniformly colored (no patterns), and circular (no tilted views)?

[Julia M. C.]

Not really. Some of them have words on them and some of my pictures are more similar to an ellipse than to a circle. These are some others examples.

Answer 1

[Image Analyst]

See http://blogs.mathworks.com/pick/2008/05/23/detecting-circles-in-an-image/?s_tid=answers_rc2-1_p4 by Brett Shoelson

Answer 2

[Image Analyst]

Since your yoyo images could be any color and any pattern over any background, this is an extremely challenging Ph.D.-level project you have chosen. You need to search the CBIR literature to see how it's done.

The only suggestion I would have is this paper on detecting ellipses in images: https://www.ecse.rpi.edu/homepages/qji/Papers/ellipse_det_icpr02.pdf