Grayscale Image Matting and Colorization

Tongbo Chen     Yan Wang     Volker Schillings     Christoph Meinel    

Abstract

This paper presents a novel approach to grayscale image matting and colorization. The first part of this approach is an efficient grayscale image matting algorithm in Bayesian framework. The foreground and background color distributions, and the alpha's distribution are modelled with spatially varying sets of Gaussians. The major novelties of this matting algorithm are the introduction of alpha's distribution and gradient into the Bayesian framework and an efficient optimization scheme. This grayscale image matting algorithm can effectively handle objects with intricate and vision sensitive boundaries, such as hair strands or facial organs. In the second part, by combining the grayscale image matting algorithm with color transferring techniques, an efficient colorization scheme is proposed, which provides great improvement over existing techniques for some difficult cases, such as human faces or images with confusing luminance distribution.

Paper

Tongbo Chen, Yan Wang, Volker Schillings, and Christoph Meinel. Grayscale image matting and colorization. In proceedings of Asian Conference on Computer Vision (ACCV 2004), Jeju Island, Korea, Jan. 27-30, 2004, pp. 1164-1169. PDF

---