3 SU subunit sign modelling for continuous sign language recognition

Abstract

This dissertation presents a 3 SU subunit sign modelling framework for automatic sign language recognition from continuous video sequences consisting of isolated words and signed sentences This thesis addresses three major issues of automatic sign language recognition namely i the importance of discriminative feature extraction and selection ii handling epentheses movements and segmentation ambiguities and iii automatic recognition of large vocabulary sign sentences and signer adaptation in a single subunit sign modelling framework Hence this thesis focuses on the three major contributions for addressing the above issues The first contribution presented in this thesis is the feature extraction and selection technique to identify the best features for recognizing the signs Hence a novel Bi fold Bayesian Parallel Hidden Markov Model BPaHMM is introduced to classify the manual and non manual features of sign languages from continuous video sequences A discriminative feature selection strategy based on the information gain is introduced to select the important features and to reduce the feature space and computation time newline