Cellular neural network for the extraction of linked visual features

Important features of a 3D scene are mapped into the resulting 2D image as sharp illuminance variations. These edgels (edge elements) are detected by first-order differentiation operator and linked together with edgels in the vicinity to form the primal sketch on a basis of orientation continuity. Linked edgels are more representative of 3D features and attenuate the presence of noisy isolated edgels. Cellular neural networks (CNN) offer many advantages in vision-based applications. The CNN's local processing feature is well adapted to vision algorithms and facilitates VLSI implementation. We propose a CNN architecture using a large circular neighborhood coupled with a directional induced gradient field to link together edgels with similar and continuous orientation. The CNN is tested to extract lineaments in remote sensing images. Lineaments are long linear segments mapping large geological structures into the image.