An Active Resistor Mesh Embedding Cortical Visual Processing
In preattentive vision a vast arrays of intercommunicating, identical processes are carried out for perceiving edges by texture differences, depth mapping, computing optical flow, recovering local surface structure, segmenting images, etc. To perform these tasks in real time, analog hardware devices are required. Information is distributed, being mapped directly in the electrical variables, and computation is carried out massively in parallel with high efficiency and speed. Two steps can be distinguished when formalising these computations: 1) mapping the image onto an intermediate abstract representation by means of appropriate receptive fields and 2) arranging information in functional maps, thus generating useful image descriptors for subsequent processing. Relating these computations to the signal processing capabilities of active resistor meshes, may ensure the most efficient use of hardware. In this paper we extend the approach of  to not homogeneous anisotropic meshes. The mesh has different orientation-selective neurons organized on the same layer, to emulate the orientation domains in the mammalian visual cortex. In a discrete model, the orientation map is defined as a 2D array in which every node is associated with a preferred orientation ranging from 0 to π radians . Such orientation maps allow the fusion of information from different channels.