Clothing segmentation using foreground and background estimation based on the constrained Delaunay triangulation

This paper proposes a new clothing segmentation method using foreground (clothing) and background (non-clothing) estimation based on the constrained Delaunay triangulation (CDT), without any pre-defined clothing model. In our method, the clothing is extracted by graph cuts, where the foreground seeds and background seeds are determined automatically. The foreground seeds are found by torso detection based on dominant colors determination, and the background seeds are estimated based on CDT. With the determined seeds, the color distributions of the foreground and background are modeled by Gaussian mixture models and filtered by a CDT-based noise suppression algorithm for more robust and accurate segmentation. Experimental results show that our clothing segmentation method is able to extract different clothing from static images with variations in backgrounds and lighting conditions.