The human visual system can determine the distance to an object from its retinal images if the system "knows" which point in each image corresponds to a given point in the object. Our model solves this "correspondence problem" by representing an area of visual cortex onto which images are "printed" as a pattern of neuronal excitations on a 2-dimensional array of identical interconnected model neurons specified by activation level, activation function, decay constant, connection weights, synaptic depression and recovery, and thresholds. As this system evolves with time the dynamics of the spreading neuronal excitation pattern solve the correspondence problem using mechanisms similar to those we used previously to detect moving objects. Propagation in the cortical direction parallel to the interpupillary line is used to superimpose the two retinotopic cortical images so that their fixation points coincide. Under these conditions corresponding points lie very close to each other. The model neuron excited by the superimposed fixation points becomes "superexcited" and sends waves of excitation in both directions along the interpupillary line. These waves encounter corresponding points in close sequence as paired neighbors with no intervening points. Crowded images with large disparities may yield incorrect pairings, perhaps corresponding to Panum's fusional area. Pairs excited by impinging wave fronts bind to each other and project to a parallel cortical layer where the original translation process is reversed and the bound pairs assure the correct pairing of the original retinal image points. This process operates simultaneously on all the "rows" of neurons parallel to the interpupillary line and yields a complete depth description of a 3-dimensional visual surface in very few steps.
ASJC Scopus subject areas
- Sensory Systems