作者: Shifeng Li , Hu-Chuan Lu , Xiang Ruan , Yen-Wei Chen
DOI: 10.1007/S10044-011-0220-3
关键词: Pattern recognition (psychology) 、 Torso 、 Basis (linear algebra) 、 Segmentation 、 Scale (ratio) 、 Probabilistic logic 、 Artificial intelligence 、 Computer science 、 Computer vision 、 Cut 、 Human body
摘要: In this paper, we propose a novel method to segment human body in static images by graph cuts based on two deformable models at two-scale superpixel. our study, segmentation is decomposed into torso detection and lower recovery. Based the first-scale superpixel, seeds of are obtained basis coarse region, which estimated an improved model. For body, estimate hip region obtain second-scale Besides, upper leg model designed derive more foreground body. To avoid failure caused heavy dependence between hierarchies, scheme probabilistic hierarchical presented. Experiments datasets containing 200 photographed ourselves 100 other collected from public show that approach can accurately with variety poses, backgrounds clothing. Segmenting image