Interactive Graph Cuts using FPGA

In this paper, we describe an FPGA system for interactive graph cuts. With this system, a whole image is divided into two regions, objects and their background, based on the regions specified as the objects or background by the user. This procedure can also be applied to video sequences. For the division, a directed graph is generated for each image, the graph is divided using the max-flow algorithm and only the objects are shown to the user. For accelerating these procedures which require irregular memory accesses in nature using an FPGA, efficient data management methods are required. By caching a part of the image data, and by processing the graph on a deeply pipelined circuit, the system achieved 20-30 fps for standard benchmark images.