Multiscale Wavelet Feature Extraction Integrated with CNN for Improved Gesture Prediction

Abstract This study focuses on the application of multiscale wavelet analysis to hand gesture recognition. A crucial component of human-computer interaction, hand gesture recognition makes it possible for a variety of applications, including assistive technology, robotics, and virtual reality, to have natural and intuitive communication interfaces. For reliable and accurate hand gesture detection, this study proposes a hybrid method that combines convolutional neural networks (CNNs) and multiscale wavelet transforms. The system initially breaks down gesture images into several frequency bands using 2D wavelet transforms, thereby capturing both high-level and low-level data. In order to extract discriminative features across scales and orientations, several wavelet families—such as Haar, Daubechies, Coiflets, and Morlet—are investigated. These wavelet-based features are then fed into a custom CNN architecture for deep learning-based classification. The proposed model is evaluated on a benchmark hand gesture dataset and demonstrates superior performance compared to traditional image-based methods. The results indicate improved accuracy, noise robustness, and generalization across gesture types and users. This study highlights the potential of combining multiscale signal analysis with deep learning for scalable and real-time gesture recognition systems.