International Journal of Engineering and Technology

This research paper highlights the use of shape and texture information for recognizing gestures of Indian sign language. The proposed method involves extracting the hand segments from the original color gesture images and subjecting them to further processing. In the next stage texture information of the hands in extracted using gabor filter. Again from the segmented hand portions shape is modeled using Chan-Vese(CV) active contour model. Finally both the shape and texture information are merged together to produce a feature vector that essentially represents a sign in Indian Sign Language. To reduce the dimensionality of the feature matrix principle component analysis is applied on the feature matrix. The obtained feature matrix will train a artificial neural network the learns using error back propagation algorithm. Indian sign language database was created for around 36 signs with 10 different signers. For training 4 sets gesture images were used and the remaining 6 sets were used for testing. After extensive testing under various conditions the average recognition rate stands at 98.2%.