A message passing algorithm for MRF inference with unknown graphs and its applications

Abstract

Recent research shows that estimating labels and graph structures simultaneously in Markov random Fields can be achieved via solving LP problems. The scalability is a bottleneck that prevents applying such technique to larger problems such as image segmentation and object detection. Here we present a fast message passing algorithm based on the mixed-integer bilinear programming formulation of the original problem. We apply our algorithm to both synthetic data and real-world applications. It compares favourably with previous methods.