Indian Alphabets and Digits Sign Recognition Using Pretrained Model

Abstract

Deaf community uses sign language as their natural language and also as their first language. Indian sign language (ISL) is primary language for deaf people of India, which they use for interacting with other people. Also, not all deaf are aware with reading and writing which makes communication even harder for them. The automated sign language recognition system comes as necessity when human interpreters are not available in urgent situations. In India, local sign languages are used to ease communication among deaf, which lacks the use of standardized sign language. It is important to note that each country has its own unique sign language. Indian Sign Language Research and Training Center (ISLRTC) has launched the first ISL dictionary that can be used as national sign language of India. This paper proposes an approach that uses a novel dataset created using signs of alphabets and digits by ISLRTC to recognize images of ISL. Moreover, the approach uses transfer learning that gives good accuracy with limited number of ISL signs. We have achieved 97.26% accuracy using MobileNet as pretrained model using our approach on our own dataset.