Automatic trimap and alpha-matte generation for digital image matting

Natural image matting is the process of softly fetching foreground object from the image. A crucial issue of Image Matting is to provide trimap manually, which is tedious and time consuming job for user. A carefully specified trimap is needed as the accuracy of trimap-based approach reduces drastically as we move from finer to coarse trimaps. Although scribble-based approaches reduced user burden, they compromised with the accuracy. In this paper we propose an approach to automatically generate trimap and then refine the trimap using region growing mechanism to give fine and accurate trimap. Afterwards it computes alpha-matte through Learning Based Matting using this fine trimap to give high quality alpha-matte in very less time. It helps in obtaining accurate result without user intervention in very less time. We demonstrate experimentally that automatically generated trimap and alpha-matte is not only qualitatively and quantitatively accurate but also faster than scribble-based approach.