ASL Classification System Using Hierarchical Features Based on appearance, moment invariants, and direction

An American sign language(ASL) clssification system using hierarchical features based on appearance, moment invariants, and direction is proposed. First, to reduce the error rate and illumination chage, infrared image is used in this article. The appearance feature calculates circularity and eccentricity. And then ASL is classified by K-means using them. Next, the Hu- moment invariants features are used to recognize ASL by back-propagation (BP). Finally, the direction feature can accurately classify similar gestures like G and Z, I and J, U and H. The goal of this article is to achieve an efficient and effective classification system that meets the accurate classification of gestures. As a result, the performance evaluations of the recognition rate shows 97.15% and it takes 0.046 s to process one frame.