In sign language systems, facial expressions are an intrinsic component that usually accompanies hand gestures. The facial expressions would modify or change the meaning of hand gesture into a statement, a question or improve the meaning and understanding of hand gestures. The scientific literature available in Indian Sign Language (ISL) on facial expression recognition is scanty. Contrary to American Sign Language (ASL), head movements are less conspicuous in ISL and the answers to questions such as yes or no are signed by hand. Purpose of this paper is to present our work in recognizing facial expression changes in isolated ISL sentences. Facial gesture pattern results in the change of skin textures by forming wrinkles and furrows. Gabor wavelet method is well-known for capturing subtle textural changes on surfaces. Therefore, a unique approach was developed to model facial expression changes with Gabor wavelet parameters that were chosen from partitioned face areas. These parameters were incorporated with Euclidian distance measure. Multi class SVM classifier was used in this recognition system to identify facial expressions in an isolated facial expression sequences in ISL. An accuracy of 92.12 % was achieved by our proposed system.
展开▼
