PCA AND PROJECTION BASED HAND GESTURE RECOGNITION

Developing hand gesture recognition algorithms, and more generally, pattern recognition algorithms is a very active area of research in computer vision. There are various approaches and techniques to the recognition problem among researchers. In this manuscript, our objective is to develop a novel Principal Component Analysis based hand gesture recognition algorithm, and compare its performance against k-Nearest Neighbor classifier and Sparse Representation based Classifier. The proposed algorithm makes use of linear triplet loss embedding and projections onto subspaces. An open source HandReader dataset consisting of 500 labeled images with 10 signs from American Sign Language is split into a training set with 100 images and a test set with 400 images. The proposed algorithm outperforms with 95% accuracy. This shows that the proposal methodology might be effective in computer vision when there is relatively small amount of data is available. It is expected that approaches similar to the current one will contribute the emergence of machine learning algorithms with Principal Component Analysis based techniques.