Ant colony alpha matte: A new approach for natural image matting

In this paper, we present an interactive algorithm to separate foreground and background regions of natural images (natural image matting) using ant colony optimization. Today, image matting is one of the most challenging and interesting research fields in image processing. In our approach instead of preparing a trimap, the user specifies foreground and background regions by some red and blue scribbles. Then by minimizing local energy function of all pixels alpha matte is estimated. Our approach not only needs a little interaction with the user but also by applying ant colony algorithm on color images, finds homogenous regions of the image and yields good results compared with other methods. In other words, the local energy of a pixel is obtained using traveled path by the pixel ant and since the ant tends to move to pixels similar to beginning pixel, homogenous regions of the image are detected. Moreover, we use some techniques like vectorization in the implementation of our algorithm in order to decrease time complexity. Experimental results show our algorithm advantages.