Local Intensity Directional Boosted Hand Gesture Recognition using Machine Learning Technique

This research work aims to recognize hand gestures by the extraction of local features from hand sign images. In recent times, the recognition has had a high level of attraction in the communication of physically challenged people. The local feature of visual information from an image involves texture, which uses properties such as pixel intensities or local directionality for edge details. These properties are used to analyze the structural arrangement of pixels information in hand signs. However, texture-based feature descriptors do not contemplate the two properties simultaneously in the recognition system. In this paper, modified pixel intensities and local directional patterns have been used simultaneously to enrich the discriminant features of an image for better recognition. The proposed work modified pixel intensities to analyze the maximum and minimum intensities of hand gesture images. Furthermore, modified pixel intensities are integrated with local directional patterns to obtain new descriptors, namely for robust distinct features. The experimental results were validated using various datasets such as Jochen Triesch of uniform light, dark background, National University of Singapore (NUS-I) and ASL-alphabets kaggle by Support Vector Machine (SVM) and K-nearest neighbor (K-NN) classifier. The estimated recognition rates of various datasets are compared with existing techniques. The proposed work reflects high performance when compared to state-of-the-art descriptors.