Representation of eigencurves based on various clothes

This paper investigates an influence of human dresses, called dress effect, on an eigenspace generating. To escape from this effect, we propose to employ a blurred edge image of the original. This paper also proposes a mean appearance-based approach for recognizing human postures, which considers some sort of appearance-change due to various dress environments. We have obtained promising performance on human posture recognition using the proposed method.