Improved Convolutional Neural Network based Sign Language Recognition

Abstract

Hand gestures offer a natural manner for humans to act with computers to perform a spread of various applications. However, factors like the quality of hand gesture structures, variations in hand size, hand posture, and environmental illumination will influence the performance of linguistic communication recognition algorithms. Recent advances in Deep Learning have considerably advanced the performance of image recognition systems. particularly, the Improved Convolutional Neural Network has incontestable superior performance in image illustration and classification, compared to standard machine learning approaches. This paper proposes AN Improved Convolutional Neural Network (ICNN) appropriate for linguistic communication recognition tasks. information augmentation is at the start applied that shifts pictures each horizontally and vertically to an extent of 200th of the first dimensions every which way, in order to numerically increase the scale of the dataset and to feature the lustiness required for a deep learning approach. These pictures area unit input into the projected ICNN model that is authorised by the presence of network low-level formatting (ReLU and Softmax) and L2 Regularization to eliminate the matter of information overfitting. With these modifications, the experimental results mistreatment the ICNN model demonstrate that it's a good methodology of accelerating the performance of CNN for linguistic communication recognition. The model was trained and tested mistreatment 105600 static hand gesture pictures, that incorporate variations in options like scale, rotation, translation, illumination and noise. The projected ICNN was compared to a baseline Convolutional Neural Network and also the results show that the projected ICNN achieved a classification recognition accuracy of 99.96.