I have this image:
What I would like to do is classify this image between the flowers and trees, so that I could find the region of the image that is covered by trees, and the region that is covered by those flowers.
I was thinking that this could be some kind of FFT problem, but I'm not exactly sure how it would work. The FFT of the individual flower is different that the trees, so I could compare magnitudes there or something, but I dont know if thats the exact right approach.
The reason I was leaning down this route is because I have, in the past, written an image classification algorithm that relied on magnitude data to distinguish different areas of an image, but I'm ju开发者_如何学Gost not sure how to generate that, or if its the right approach.
Thanks for any tips
You might try texture-based approaches such as o co-occurence matrix. Reasonably close to your FFT approach (you look for patterns in local similarity), but not restricted to simple frequencies.
What if you try extracting color planes from the RGB image? The "greener" components (i.e. the trees) should lie all in the green plane in RGB color space, whereas flowers will share components between red, green and blue (thus if you average the three planes I expect you to see the flowers enhanced.
精彩评论