A NEURAL-NETWORK APPROACH TO THE DETECTION OF TEXTURE BOUNDARIES

A neural-network architecture is used here to find boundaries between differently textured surfaces in an image. The architecture is trained on boundaries between a specific set of textures and is shown to be a ble to generalize its training to other textures which were not part o f the training set. This is a first step towards the construction of a general texture boundary-detection system which would not require any information concerning the textures involved. Co-occurrence measures are used as a texture representation, and the cascade-correlation arch itecture is employed as the neural-network classifier.