A Comparative Analysis of Filters towards Sign Language Recognition

Abstract

One of the major drawbacks of our society is the barrier that is created between the deaf and the normal. Communication is the only medium by which we can share our message through speech. But, for a person who is having hearing impairment faces difficulties in communication with normal people, sign language is the main medium for deaf people to communicate. There have been few attempts in the past to recognize hand gestures, but with low rate of recognition. Furthermore, the dataset accessible in this subject is still noisy. The objective of this study is to determine filtering techniques that give good performance for hand sign language recognition. Dataset used in this study is Gesture Dataset 2012. In this study, three types of filtering techniques have been chosen which are average filter, median filter, and Laws’ masks filters. A comparative analysis has been done and the performance of the filtering techniques are based on recognition performance. Meanwhile, Canny, Sobel, Prewitt, and Roberts algorithms are used for edge detection and features extracted are Histogram of Oriented Gradients (HOG). Finally, for classification, multiclass Support Vector Machine (SVM) classifier is used. At the end of this study, it has been found that using a combination of Laws’ masks filters and Canny edge detection algorithm is proven to be a promising combination to increase sign language recognition performance since they give the best results of recognition performance compared to other methods used in the experiments.