Adaptive skin colour modelling for hand and face segmentation

Abstract

A real-time skin colour segmentation is a prime concern in posture/hand gesture based sign language recognition. The developed novel method separate posture/hand gesture without any marker on any background by taking into account foreground skin colour. The proposed algorithm, segmenting dynamic posture/hand gesture done in front of the camera or from images or video database. The values of YCgCr are determined in real time and form YCgCr adaptive bound. These bound values are used to segment foreground skin colour from skin / non-skin colour background. The algorithm deployed on the embedded image processing hardware platform and any skin colour/tone it can adapt and used further to segment face/ hand gestures. The proposed adaptive skin colour algorithm accuracy for standard Ali Yawar Jung video dataset of ISL hand gestures is 92.64% and visually when two different signers performed same gestures the average accuracy is 91.82%.