Improvements of Bayesian Matting
Mikhail Sindeyev, Конушин Вадим Сергеевич, Vladimir Vezhnevets · 2007
Digital image matting is a process of extracting a foreground object from an arbitrary natural image. Unlike the image segmentation task it is required to process fuzzy objects (like hair, feathers, etc.) and produce correct opacity channel for them. The result can then be composited onto a new background or edited by processing foreground and background layers separately. Digital image matting has become a compulsory step in many photo-editing and video-compositing tasks. Currently professional digital artists have to accurately trace objects contours and paint the details to achieve maximum quality. Our aim is to create a convenient workflow for automating this process and make it possible to effectively handle high-resolution images. In this paper we show how a smoothness constraint can be incorporated into Bayesian matting algorithm framework as additional regularization to improve the result quality without affecting the computation speed. We also demonstrate the hierarchical approach that significantly increases processing speed without noticeable loss of quality. This allows us to create convenient digital image matting system.