Human segmentation with deep contour-aware network

Human detection and segmenting are important computer vision problems with applications in indexing, surveillance, 3D reconstruction and action recognition. The figure-ground segmentation of humans in images captured in real-world environment is a challenge problem due to a variety of viewpoints, articulated skeletal structure, complex backgrounds, varying body proportions and clothing, etc. In this paper, we proposed a new approach to human segmentation in still images based on Deep Contour-Aware Network (DCAN), which is a unified multi-task deep learning framework combining the complementary object and contour information simultaneously for better segmentation performance. Experimental results on a large-scale human dataset indicates our human segmentation method can achieve a marginally better segmentation accuracy than the state of the art works.