Evaluation of Dilated CNN for Hand Gesture Classification

Abstract

Convolutional neural networks (CNNs) have been widely used in hand gesture classification problems, and have made a major contribution to this area by overcoming the limitations of hard-code feature extraction techniques. CNN in hand gesture classification aims to improve performance through automatic feature engineering. Several researchers have used various CNN architectures to accurately classify hand gestures.In this paper, we investigate the performance of a popular CNN variant called dilated CNN to classify hand gestures into their corresponding classes. We compared the performance of the dilated CNN with that of the standard CNN on two benchmark ISL and ASL datasets. The experimental results demonstrate that the dilated CNN significantly enhances performance compared to the standard CNN. We obtained a significant increase in accuracy for both datasets using the dilated-CNN compared to the standard CNN.