Appearance Based Recognition Methodology for Recognising Fingerspelling Alphabets

摘要

In this paper, a study on the suitability of an appearance based model, specifically PCA based model, for the purpose of recognising fingerspelling (sign language) alphabets is made. Its recognition performance on a large and varied real time dataset is analysed. In order to enhance the performance of a PCA based model, we suggest to incorporate a sort of pre-processing operation both during training and recognition. An exhaustive experiment conducted on a large number of fingerspelling alphabet images taken from 20 different individuals in real environment has revealed that the suggested pre-processing has a drastic impact in improving the performance of a conventional PCA based model.