Distinguishing line-detection from texture segregation using amodular network-based model

摘要

In order to enable task-dependent processing of lines, edges, andtextures, the authors introduce a network-based model for the front-endvisual system. The purpose of this model is to provide differentrepresentations of a retinal image in such a way that different actionsand decisions about the presence of objects in the visual scene can beundertaken at a further stage. In particular, as a starting point, themodel with distinguish lines or edges from textures. A hierarchy of ANN(artificial neural network) modules is introduced for performing twoessentially different tasks: line and edge detection and texturesegregation. The network module is the Entropy Driven Artificial NeuralNetwork module. The model has been implemented on a parallel computer, a16-processor MEIKO transputer system. The model's ability to distinguishline and edge detection from texture segregation, starting from the samebank of Gabor filters, is demonstrated