A Statistical Approach to Efficient Curvature Scale Space Matching for Recognizing Hand Gestures

In the paper, we develop an efficient matching algorithm for recognizing different types of hand gestures. The algorithm has been performed on an input image window in three stages: first the skin regions are segmented using a global threshold technique. Then a curvature scale space (CSS) image is created considering the largest contour of the segmented skin regions. Finally, a novel approach using statistical measure has been applied to match the input CSS image and the set of previously stored model CSS images. The algorithm is robust since it uses global distribution of the CSS image as part of the matching algorithm and performs better than the previous methods applied for shape similarity using curvature scale space (CSS) matching.