A greedy clustering algorithm along the time axis for Chinese language recognition

Abstract

Sign language is the language used by the deaf. Chinese Sign Language (CSL) is primarily composed of a gesture language complemented by a finger-spelling language, containing 30 finger-spelling alphabets and about 5,500 basic sign words. This paper focuses on CSL recognition. Using the characteristics of the sign words in the CSL gesture language, an algorithm called GLATA (Greedy cLustering Algorithm along the Time Axis) is proposed to segment the sign words, a training algorithm based on GLATA is proposed to train the template corresponding to each word, and a recognition algorithm based on GLATA is used to recognize 227 words randomly selected from CSL with an accuracy of 96%. In addition, a sign segmentation technique based on the overall speed of the hand is proposed to delete movement epenthesis in continuous signing. From the experimental results, it is shown that GLATA is an effective algorithm for CSL recognition.