Transfer-learning analysis for sign language classification models

Abstract

Alphabet sign language recognition with high accuracy is a tedious task in computer vision due to several reasons like the lack of sufficient quantities of the annotated dataset, signer variety, continuous signing, etc. The relative scarcity of labelled data in the sign language recognition field has impeded the exploitation of the different models. For the above-mentioned problems, the researchers perform other augmentation techniques. In this paper, we evaluate four different transfer learning approach on three publically available datasets. The various transfer learning models explore the influence of different data augmentation techniques that should increase the performance of the classification model. All the four model makes a comparison with data augmentation and without data augmentation. This study suggested that the transfer learning can improve the classification accuracy.