Correlating Edge with Parsing for Human Parsing

Human parsing has great application prospects in the field of computer vision, but there are still many problems. In existing algorithms, problems small-scale target location and problem background occlusion have not been fully resolved, which will lead to wrong segmentation or incomplete segmentation. Compared with practice feature concatenation, using correlation between two factors can make full use edge information for refined parsing. This paper proposes mechanism network (MCEP), uses spatial aware max-pooling (SMP) module capture correlation. The structure mainly includes steps, namely (1) collection operation, where, through mutual promotion features features, more attention is paid region interest around human body, clues body collected adaptively, (2) filtering where parallel adopted solve problem. Meanwhile, semantic context extraction capability endowed enhance prevent small detail loss. Through a large number experiments on multiple single-person multi-person datasets, this method greater advantages.