An FPGA Implementation of Robust Matting

Abstract

Matting is a process of extracting a foreground from background in an image. It is one of the key techniques in many image editing, and many algorithms have been proposed. Robust Matting is one of the powerful matting algorithms. In the Robust Matting, first, a set of pixels in the foreground and background are sampled, and then, for each pixel in the image, its category is determined by comparing it to a number of pairs of a foreground and background sample. Its computational complexity is very high because of a large number of multiply and square operations. In this paper, we propose an FPGA implementation for real-time processing of HD images. In our approach, the image is divided into small blocks and the same pairs of the foreground and background pixels are used for the pixels in the same block in order to reduce the number of the operations. This approach makes it possible to execute multiply operations by simply looking up tables that are dynamically updated block by block.