Sign Language Recognition using CLBP and DWT features with KPCA-based Feature Reduction Approach
Dimpy Sahu, Suvendu Rup · 2024
Sign language recognition (SLR) is an active area of research in the field of Computer Vision. Sign languages are visual languages that communicate using the hands, facial emotion and the body movements. Sign Language serves as a bridge between us and the individuals with impaired hearing or linguistic activities. It allows individuals to understand the world around them through visual descriptions, and hence contribute to society. The present work aims at developing an efficient SLR model that can be classified multi-class sign languages effectively. First, a combined feature extractor strategy is suggested by employing Discrete wavelet transform (DWT) and Circular local binary pattern(CLBP) to extract the directional and texture features to construct a robust feature descriptor. Subsequently, Kernel Principal Component (KPCA) analysis is utilised for the purpose of feature reduction. For classification, three different classifiers namely SVM, KNN and Logistic Regression are suggested as the combination of DWT and LBP features, KPCA based feature reduction and classification. Out of these combination DWT+CLBP+KPCA+SVM attains a maximum accuracy of $\mathbf{9 8. 7 \%}$ for American Sign Language(ASL) and $\mathbf{9 6. 4 \%}$ for Indian Sign Language dataset. From experimental results and analysis it has been observed that the proposed SLR model shows superior performance as compared to other benchmark schemes.