Automatic Foreground Segmentation Using Light Field Images

Foreground segmentation is a fundamental method in computer vision. Traditional foreground segmentation algorithms are sensitive to blurry degree of background, smooth foreground regions and camouflage foreground. To deal with these problems, we use light field images as input by exploiting its focusness cue. In this paper, we propose an automatic foreground segmentation algorithm for light field images. Firstly we calculate focusness measures in refocused stack and over-segment all focus images. Then the graph cut framework is introduced to generate foreground result. Experiments show that our method has a superior performance against counterparts and is more robust in scenes like ambiguous foreground.