Action-based neural networks for effective recognition of images

This paper presents a novel approach to the recognition of images or scenes, by associating human perception actions to the system and introducing neural network architectures that are able to learn the derived representations. The approach is related to recent research efforts towards a deeper understanding of human information processing and uses appropriate recurrent neural networks for generating the desired associations in time varying environments. Initial results obtained when applying the proposed approach to the problem of recognition of images of objects, that are deformed and/or corrupted by noise, are very encouraging