A methodology of extracting standing human bodies from single images.

摘要

Extraction of human bodies in images is a difficult task that facilitates numerous applications, like the understanding of skin and recognition of activity. In order to come up with a highly dimensional pose space and complexity of scene, a template matching process is required. The majority of this work requires complex training and template matching techniques.;A bottom up methodology is proposed for automatic extraction of standing human bodies from single images in case of almost upright poses in a cluttered environment. Color, position and face are used to identify and localize the human body .The segmentation of different levels is combined to extract the pose with the highest potential. The human body segments arise through the joint estimation of foreground and background during the search of body parts, which provides exact shape matching.