Transform-based Arabic sign language recognition

Sign language is an independent language that uses gestures and body language to convey meaning. Sign language recognition facilities the communication between deaf and community. In this paper, we investigated the use of different transformation techniques for extraction and description of features from an accumulation of signs' frames into a single image. We show the performance of three transformation techniques (viz. Fourier, Hartley, and Log-Gabor transforms) applied on the whole and slices of the accumulated sign's image. In addition, different classification schemes are tested and compared. Overall system's accuracy reached over 99% for Hartley transform which is comparable with other works using the same dataset.