Natural Image Matting Using Cellular Automata

This paper presents an interactive algorithm for natural image matting. Given a few user-marked foreground and background strokes, the unmarked region of the input image is modeled as a Markov Random Field (MRF) and its energy function is formulated as the sum of the local energy of the unmarked pixels. The opacity matte is extracted by constructing a cellular automaton on the image, of which each cell (pixel) iteratively estimates its local opacity by minimizing its local energy until convergence. The initial state of the cellular automaton is set by the user-supplied strokes and each cell's state can be changed by additional user input during the matting process, that is, user input can be added to guide the matting process without rerunning the algorithm. Experiments on many complex natural images demonstrate that visual plausible mattes can be obtained with modest user effort.