Symmetry axis extraction by a neural network

This paper proposes a neural network model that extracts axes of symmetry from visual patterns. The input patterns can be line drawings, plane figures or gray-scaled natural images taken by CCD cameras. The model is a hierarchical multi-layered network, which consists of a contrast-extracting layer, edgeextracting layers (simple and complex types), and layers extracting symmetry axes. Its architecture resembles that of the lower stages of the neocognitron. The model extracts oriented edges from the input image first, and then tries to extract axes of symmetry. To reduce the computational cost, the model checks conditions of symmetry, not directly from the oriented edges, but from a blurred version of them. The use of blurred signals endows the network with a large tolerance to deformation of input patterns, too. It is important to get blurred signals, not directly from the input image, but from the oriented edges. If the input image is directly blurred, most of the important features in the image will be lost. Since the model uses oriented edges, however, most of the important features can still remain even after the blurring operations, by which information of edge locations becomes ambiguous.