Error detection and correction approach for Arabic sign language recognition

Sign language recognition is a promising application that breaks the barrier between the deaf and normal people. However, to make it a practical application it needs to exist in a free environment. Previous developed systems suffer from the "controlled-environment" constraint, if one of the control assumptions is violated, the recognition accuracy dramatically decreases. In this paper, a post processing module based on Natural Language Processing rules is proposed to detect and correct expected errors resulting from recognition system. We suggest a semantic-oriented approach which can correct semantic level errors as well as lexical errors, and is more accurate especially for domain-specific sign language recognition error detection and correction. Through experiments, it will demonstrate the better performance of the proposed post processing approach. The experiments were done on Arabic sign language recognition and it can be generalized to other sign languages.