A Computational Framework for Indian Sign Language Recognition

Abstract

Sign language is the primary means of communication for the hard to hear and speak people around the globe. Sign language emphasizes on visual possibilities as the participants are unable to hear sound patterns. Sign language uses different signs, body postures and gestures as opposed to sound patterns for communication, and evolves like any other spoken language. American Sign Language (ASL), British sign language (BSL), Arabic sign language (ArSL), Chinese sign language (CSL) and Indian sign language (ISL) are some of the widely used sign language systems around the world. There exists significant variation between sign languages, and due to these inherent variations, it is not possible to fully adopt a methodology that is found suitable for all. There are enormous complexities in ISL. Contrary to ASL, ISL sentences follow Subject-Object-Verb pattern. For example, the relative positioning of hand on face with respect to nose can convey ‘WOMAN’ or ‘THINK’ in ISL. Such complexities necessitate independent research in ISL. Sign language recognition involves integration of different categories of signs. The signs can be mainly categorized into three groups like static hand gestures, dynamic gestures and facial expression. This research focuses on these three different channels and work to identify the potential of different computational methods to address some of the associated complexities with each channel. These complexities include static gestures with resemblances, static overlaid gestures, differential movement and directional changes in dynamic gestures and facial expression changes.