Patch-based finger encoding for hand posture recogntion

This paper proposes a method for hand posture recognition based on finger characteristics. These characteristics can be classified into fully extended or bended finger which are useful for further recognition process. Each finger is encoded for five elements codeword and each element is set up from the area of finger phalanges. This system will be decided whether each finger is extended or bended. After obtaining the finger state, the combinations of bended state (0´s) and extended state (1´s) from five fingers define the sign pattern codeword. Hamming distance between sign pattern codeword and template codeword is applied to recognize the corresponding alphabet. The result of experimenting 360 samples from nine alphabet patterns shows that the performance evaluation is about 90.83 %.