Alpha matting using robust color sampling and fully connected conditional random fields

Abstract

Alpha matting refers to the problem of softly extracting the foreground from a given image. Previous matting approaches often focused on using na < ve color sampling methods to estimate foreground and background colors for unknown pixels. Existing sampling-based matting methods often collect samples only near the unknown pixels, which may yield poor results if the true foreground and background samples are not found. In this paper, we present novel approach to extract foreground elements from an image through color and opacity (i.e., alpha) estimations, which consider available samples in a search window of variable size for each unknown pixel. Our proposed sampling method is robust in that similar sampling results can be generated for input trimaps of different unknown regions. Further, after the initial estimation of the alpha matte, a fully connected conditional random field (CRF) is used to correct the predicted matte at the pixel level. Our experiments show that visually plausible alpha mattes can indeed be produced.