Hand Shape Recognition from Natural Hand Position

Abstract

The article proposes a new approach to hand shape recognition that allows comparison of hands with separated fingers as well as with touching fingers. The idea is to construct a flexible hand model from the set of hand images in different poses. This model allows us to simulate different hand transformations (for example, rotations of fingers) and analyze hand shape after them. Person recognition is performed by matching the reference model to a test hand image. The proposed similarity measure is normalized symmetric difference between superposed silhouettes of test hand and transformed reference hand. The article also introduces a~prototype of peg-free person recognition system based on the proposed approach and experimental results on real data.