Alternating Scheme for Supervised Parameter Learning with Application to Image Segmentation

摘要

This paper presents a novel alternating scheme for supervised parameter learning. While in previous methods parameters were optimized simultaneously, we propose to optimize parameters in an alternating way. In doing so the computational amount is reduced significantly. The method is applied to four image segmentation algorithms and compared with exhaustive search and a coarse-to-fine approach. The results show the efficiency of the proposed scheme.