Using Gabor filter bank with downsampling and SVM for visual sign language alphabet recognition

With the increasing advances in computer vision, the research on automated human gesture and sign language has attracted the attention of many researchers. It has many applications for human-computer interaction helping persons with hearing impairment in smart environments. In this paper, we focus on static hand visual features to build a system for recognizing hand and finger gestures representing different sign language alphabets. After hand segmentation, the proposed method employs texture based features extracted by down-sampling Gabor-transformed images using multiple scales and orientations. Then, a support vector machine is used for multi-class classification. The evaluation of the proposed approach on a benchmark dataset for the American sign language has reported over 95% overall accuracy with several signs perfectly recognized.