Automatic feature extraction from front and side images

This paper presents a systematic method to detect feature points on the silhouette of human body from front and side images. Based on the correspondence between the coordinates and the eight numbering system of chain codes, the codes that depicted at 90 degree angle were taken as corner points. The system then connects the other associated points horizontally and vertically. Thus, the human landmark positions of head, neck, shoulder, chest, waist, hip, thigh, knee, shank, ankle, and foot can be identified by the relationship between two correlative points. A total of 55 feature points and 26 body measurements can be extracted automatically. The method has been tested on four human subjects and all were correctly extracted.