A selective attention neural network for invariant recognition of distorted objects

Summary form only given. Selective attention is used to reduce the number of inputs and to recognize input scenes containing multiple objects and distorted objects at any translation or orientation. Invariance to translation and orientation is achieved by developing appropriate input representations. A recurrent object recognition network was implemented, and the network was tested with the TC problem. Using a proper input representation and encoding scheme, the networks are trained with objects in a standard position and orientation. The trained network recognizes objects at any translation and orientation and generalizes to distorted objects