Novel Indian Sign Language Recognition technique using Energy Compaction of five orthogonal transforms

Abstract

Sign language is the most basic communication medium for deaf and dumb people. It has evolved as the major area of research and study. In this paper the novel techniques for Indian Sign Language Recognition are proposed and analyzed with experimentation. Indian Sign Language has total 26 alphabets. With the help of Energy Compaction using five different orthogonal transforms, maximum energy is packed into low frequency region of the row mean of column transformed sign images. The feature vectors are extracted in five different ways from the transformed sign images in the form of feature sets of 3.125%, 6.25%, 12.5%, 25%, 50% of the total 100% coefficients of row mean of column transformed Sign images. The experimentation is done on a database containing 260 images spread across 26 categories. For each query fired on the database the average precision values are calculated. The results have improved with fractional coefficients compared to complete transformed sign image resulting in faster recognition. Overall Haar and Cosine transforms have given good results as indicated by higher precision values.