Deep Neural Networks vs Bag of Features for Hand Gesture Recognition

Deep Neural Networks and their associated learning paradigm, Deep Learning, represent one of the hottest approaches used in image understanding and recognition. As their performances depends quasi-linearly on the amount of available data, the typical case studies in the literature assume the availability of huge datasets. This paper proposes to analyze several deep neural networks (trained from the scratch or pretrained), test their efficiency in the problem of hand gesture recognition, and compare the results to a state-of-the-art classical method, the bag of features, for the case of small databases.