I am working in image segmentation and I would like to ask you something about my problem. I have a image such as bellow. I want to segment leopard from background. However, intensity information is not useful for my case because the bough (background) whose intensity distribution is similar with leopard's. Could you suggest to me some method to segment leopard? I attacted one example using active contour.Thank you in advance
1 Answer
In the first place you must admit that what you are trying to do is about impossible.
Looking at details of the image, the differences are subtle and it is very hard to tell that details belong to distinct entities.
Actually, for an algorithm there are no real "objects" to be seen but areas with different micro- and macro-texturing.
The spots on the fur are rather large and at some place you don't know if they belong to a single macro-texture or are just independent zones.
Actually, you must have previous knowledge of what a leopard is to see it. Any algorithm lacking this information will have a hard time putting the pieces together and telling the difference with the wood.
The classical approaches are by measuring texture features, such as those obtained by a bank of Gabor filters, followed by multispectral segmentation. But there is no easy roadmap, I am afraid.