Discrete Wavelet Transform on static hand gesture recognition

Abstract

The technology in present-day is very useful for human-physical recognition and the hand gesture recognition is one of them. Nevertheless, that technology still had various weakness, for example in the image brightness, contrast, recognition time, and accuracy rate. The objective of this paper was to construct hand gesture recognition system using RGB images. Pre-processing is wrought by resizing the image, separate the hand area, and pick the specific layer. This experiment used the YCbCr because its derived directly from RGB and had a higher contrast compared to other layers. The feature value was gathered from feature extraction on Discrete Wavelet Transform (DWT) using Low-Low sub-band and 2nd level decomposition. The current sub-band had smoothest contours in comparison with other sub-band. The final process was gesture classification using Hidden Markov Models (HMM) and K-Nearest Neighbor (KNN). The amount of training and testing data used were 150 and 100 images respectively, divided into five gestures with accuracy using HMM and KNN consecutively was 58% and 100%. The research novelty was that the classification impacted positively on accuracy level.