User Independent System of Hand Postures Recognition Using Part-Based Shape Representation

We present in this paper, new method for recognizing hand postures from their shapes. Firstly, convex and concave points with high curvature values are extracted from the contour of the hand´s posture shape. Then curves are fitted to the concave points. These curves are used for partitioning 2D hand´s posture shape into subparts representing the fingers and palm. Using this partitioning we define textual descriptor of hand´s posture shape. In addition we introduce a part description based on attributes consisting of relative lengths, widths and inclination angles. Recognition of postures of the hand is performed on exploiting these local descriptors and the anthropometric relations defined in [1]. The experimental results show that the proposed approach recognizes a large number of hand postures back of palm or palm faces camera (adduction, abduction, flexion and rotation). Our descriptor is a scale change invariant.