ABSTRACT

The paper aims to propose a novel technique that recognizes finger spelled American Sign Language (ASL) gestures. The external characteristic of hand, i.e. shape based algorithm is being used for recognition. Since almost all of the alphabets have a unique shape, each alphabet is characterized on the basis landmark points marked on the boundary of the hand shown by the signer. A training set is made by training several images of each alphabet and the landmark points of each alphabet which produces a 72 point descriptor are stored in database. The descriptor of test image is then matched with the ones in database. Finally, Euclidean distance classifies the test images to the recognized alphabet. By increasing the number of sampling points there was an increase in accuracy rate. Thus a 180 point descriptor results in better recognition.