Novel segmentation algorithm for hand gesture recognition

Sign language is the most important methodology using which hearing and speech impaired people can interact with the rest of the world. Conversation with hearing impaired individuals gets complicated if the listener is ignorant of sign language. Hence it becomes important to construct a bridge between these two banks. Many sign language and hand gesture recognition algorithms have been developed in the recent years, to assist people who do not have knowledge of sign language to converse with the speech impaired but very few with good results exist. One of the major concerns with respect to hand gesture recognition is segregation or segmentation of the hand and identifying the gesture. This paper explores the various possible ways of segmentation using different color spaces and models and presents the best algorithm with highest accuracy to perform the same. Various experiments were conducted for over 500 different gestures and an accuracy of around 97.4% was achieved with the segmentation algorithm selected. The algorithms were implemented in MATLAB programming language on MATLAB 7.14.0.739 build R2012a.