The Comparison of Some Hidden Markov Models for Sign Language Recognition

Abstract

Sign language is one of the most popular communication technique. While it is popular, most people don't know how to use it. The problem lies in the difficulty of how many signs there to be memorized. That's why Sign Language Recognition (SLR) is needed. Sign Language Recognition (SLR) is a very interesting area for some researcher because of the complexity of its problem especially using video-based recognition of complex sign language. That's why Sign Language Recognition (SLR) has been developed for a long time. Unfortunately, every research about Sign Language Recognition has its own limitation. Some research got low accuracy, too expensive, insufficient dataset, and many other else. Every researcher also has their own way of doing a thing but usually, the pattern in Sign Language Recognition (SLR) are dataset collection, preprocessing data, feature extraction, training, and classification. For the dataset, we will use Argentina Sign Language that provides data as video. After that, we need to preprocess it using edge detection and skin detection using the help of Contrast Adaptive Histogram Equalization (CLAHE) for image enhancement. Then we will extract its feature by its movement. At last, we will train and classify the data using some modified Hidden Markov Model. In this research, we get an accuracy of 83% by classifying 10 signs using Gaussian Hidden Markov Model (Gaussian HMM) and 72% by classifying 10 signs using Multinomial Hidden Markov Model (Multinomial HMM).