Increasing the efficiency of the Persian Sign Language system based on new preprocessing method

In this paper, a systems is presented to recognize static gesture of alphabets in Persian Sign Language (PSL). The implemented system does not need any gloves or visual marking system, and just uses images captured by camera to recognize PSL alphabets. This system contains three principal phase: preprocessing, feature extraction, and classification. Preprocessing phase includes using several preprocessing methods on the image which reduces the difference among the hand gesture in the same letter group. In the second phase, Hough Transform function is used for feature extraction from images and MLP NN is used for image classification in the third phase. Results of the paper show that in spite of applying several preprocessing methods on images, the time of neural network training is reduced. Furthermore the recognition rate of PLS improves considerably. This system is able to recognize every 37 PSL alphabet by 98.91% accuracy.