Hybrid method based on multi-feature descriptor for static sign language recognition

Sign Language Recognition is an essential research problem for enabling communication with deaf-dumb people. Sign language recognition system confronts many challenges such as complex background, illumination changes, translation, rotation, and scale problem, besides system requirements such as time of recognition, robustness, performance, and computational efficiency. This paper proposes hybridization between two strong descriptors including Histogram of Oriented Gradients (HOG) and Edge Oriented Histogram (EOH) to achieve better recognition rate with relatively low memory requirements. A new feature descriptor is used as a combined feature descriptor, which joins the advantages of each descriptor to achieve good performance. Multi-class support vector machine classifier is utilized to classify the hand gestures. Experimental results demonstrate that the proposed system gives recognition rate of 96.15 % for 1AASVM classifier and 99.23 % for 1A1SVM classifier under different hand poses and complex background with changes in lightning conditions.