Locating hands in complex images using color analysis

In a gesture-based human-computer interface, hands in successive video frames must be extracted to recognize hand gestures that correspond to various commands issued by the user. This paper presents an algorithm to locate hands by detecting skin colors. We do not use controlled lighting and the background of the user can include anything in an office environment. The hand location algorithm detects skin-colored regions in an image by using a color histogram matching technique. A skin color model histogram is constructed from the model hands shown to a camera. Using the model histogram, histogram intersection is calculated for each window of an appropriate area to generate a skin color similarity image. The image is dynamically thresholded to produce skin regions that become hand candidate regions. The candidate regions are analyzed using a priori knowledge to decide the hand region. We conducted experiments with a real-time system that inputs a sequence of images of a user. The hand location results are satisfactory such that proper traces of hand movements for hand gesture recognition are generated.