Hand Gesture Recognition Using PCA Based Deep CNN Reduced Features and SVM Classifier

Abstract

Automatic recognition of vision based static hand gesture images is a challenging task due to illumination changes, diversity in user hand shape and high inter class similarities. This paper proposes novel techniques to develop a user independent hand gesture recognition system, considering the above challenges. First, the gesture recognition performance is analyzed using proposed pre-trained AlexNet features. In this proposition the deep features are extracted from fully connected (FC) layers such as 'FC6 and 'FC7' of pre-trained AlexNet. A support vector machine (SVM) based classifier with linear kernel is used to classify gesture poses. The highest recognition accuracy is evaluated using the deep feature extracted from 'FC6 and 'FC7' independently and combination of both the feature vector with SVM classifier. Second, feature dimension of deep features are reduced using principal component analysis (PCA) based dimension reduction technique for further improvement in gesture recognition accuracy. The performance of the proposed technique is evaluated using leave-one-subject-out cross validation (LOO CV) and holdout CV test. The extensive analysis is performed on 36 American Sign Language (ASL) benchmark static hand gesture dataset using both the CV test. The experimental result shows that, the proposed technique is superior as compared to state-of-the-art techniques.