Fast Belief Propagation for Segmentation
Sungchan Park, Kyungjoong Jeong, Hong Jin Jeong · ITC-CSCC :International Technical Conference on Circuits Systems, Computers and Communications · 2007
For the segmentation, there are mainly typical cut and normalization cut among the graph based techniques which show good results due to the approximated global methods on the 2D MRF. Toward a real-time color image segmentation, we will present a high-speed and parallel fast belief propagation(FBP) algorithm which is based on the GBP’s typical cut method. At N by M 2D MRF image and T iterations, O(NMT) time complexity can be reduced to O(MT) with N Processors and it has a smaller distributed memory resources O(NT) than typical cut’s O(NM), due to the small T. With a linear systolic array structure, we can increase the computational speed by the cascaded parallel processors.