Visual Static Hand Gesture Recognition for ASL using Multi-Feature Fusion Technique

Hand gestures are the nonverbal communication medium for deaf and dumb people. These gesture symbols give independent information for sign language to understand the messages. Among numerous sign languages, American Sign Language (ASL) is the most widely used for communication and research since it’s gesture postures are expressed using a single hand only. Various novel and innovative methods are used for hand gesture identification and recognition. However, accurate recognition of hand gestures in ASL is still challenging due to the similarity among the gesture classes. A systematic approach based on multi-feature fusion is presented in this research for accurate recognition of human hand gestures. The proposed technique involves the following steps in its process pipeline: a) segmenting the hand’s region of interest from the image; b) extracting features; and c) utilising an SVM classffier to recognise motions. This work applies several texture-based features, such as histogram of oriented gradient (HOG), local binary pattern (LBP), and Gabor feature (GB) to recognize the hand gestures on the above challenges. The combination of the above features is studied and used to enhance gesture recognition performance. The proposed technique is tested and validated using several publicly available hand gesture datasets. Various parameter and ablation studies prove that the proposed HOG-GB feature fusion technique achieved better recognition performance and outperformed earlier reported gesture recognition methods.