Propagation Based Selective Sampling for Alpha Matting

Abstract

Image matting refers to the problem of foreground extraction and transparency determination of the image. It is widely used in various industries and has been studied for many years. Although various matting algorithms have been proposed, most of them are not robust enough to obtain satisfactory matting results in different regions of an image, such as smooth regions, non-uniform color distribution regions and isolated color regions. The main motivation of this thesis is to develop a new matting algorithm which can extract high quality mattes from different regions of the image. Our proposed algorithm combines color sampling and propagation methods together. Unlike previous propagation-based approaches, we proposed an adaptive local and nonlocal propagation-based approach according to the detection result of different regions in the image. Our color sampling strategy, which is based on the characteristics of the superpixel, requires much less computation cost than previous methods. Experimental results show that the proposed algorithm can effectively handle different regions in the image and obtain better matting results than previous matting algorithms.