A robust finger detection based sign language recognition using pattern recognition techniques

Sign language recognition based on finger detection is arguably the main sign language used by most dumb people. It has its own phonetics, grammar and syntax that set it apart from other sign languages. Research related to sign language (SL) is only now becoming standardized. Considering the challenge of recognizing SL, in this work a new method for recognizing SL dynamic gestures is proposed. Sign language (SL) translation systems can be used to help dumb people interact with normal people with the help of a computer. Most studies on continuous recognition of sign language are done by processing frames obtained from videos at regular/equal intervals. If a developed system is powerful enough to handle both static and dynamic motions, then it will be the best system for processing frames obtained from processing consecutive gestures. The algorithm developed for the gesture recognition system in SL formulates a vision-based approach using two-dimensional discrete sinusoidal transforms (DSTs) for image compression and self-organizing maps (SOMs), or self-organizing feature maps. Kohonen’s (SOFM) Neural Networks for Pattern Recognition, simulated in MATLAB. The system showed an accuracy rate of 91 percent.