Image understanding by neuron network-representation and operation of internal image

The authors consider a model of human internal image representation and, as an application, an image understanding by a neural network. The basic idea of the image representation is the use of intermediate products of the neural recognition system in the brain. The authors suppose it to be an internally represented image (IRI) of the outer world. It is recollected by the symbolic representation of the corresponding fact and is operated to produce the internal prospects of an operation. The process corresponds to the knowledge representation and the operation on it in artificial intelligence. According to the idea of operations on IRI, an iterative process of recognition, i.e., recollection, operation is applied to the problem of image understanding. Using the model of selective attention, an image of multiple targets with position shift and deformation is separated into the components of each target sequentially.<>