A Joint Representation Learning and Feature Modeling Approach for One-class Recognition

One-class recognition is traditionally approached either as a representation\nlearning problem or a feature modeling problem. In this work, we argue that\nboth of these approaches have their own limitations; and a more effective\nsolution can be obtained by combining the two. The proposed approach is based\non the combination of a generative framework and a one-class classification\nmethod. First, we learn generative features using the one-class data with a\ngenerative framework. We augment the learned features with the corresponding\nreconstruction errors to obtain augmented features. Then, we qualitatively\nidentify a suitable feature distribution that reduces the redundancy in the\nchosen classifier space. Finally, we force the augmented features to take the\nform of this distribution using an adversarial framework. We test the\neffectiveness of the proposed method on three one-class classification tasks\nand obtain state-of-the-art results.\n