Hand Feature Learning for Hand Shape Recognition

Hand gestures are the main elements of non-verbal communication of Deaf Community. Computer recognition of sign language deals with detecting hand gestures in videos and classifying them according to their previously assigned meanings. This project focuses on extracting hand shape features from 2D hand images cropped from sign language videos. Most of the works and researches in this report focused on auto-encoder based neural network architectures. My recent works focused on data preprocessing and comparisons between classification results by only using extracted features and results from state-of-the-art pretrained neural network classifiers by using images itself. Experiments shows that generic features which are extracted by unsupervised methods are not as sufficient as the features that are extracted by supervised classifier.