Stereovision: a model of human stereopsis

Sammanfattning: A model of the human stereopsis mechanism is presented. As foundation for the model lies a number of ideas that has arisen from redefining the correspondence problem. Instead of establishing potential matches by the detection and matching of some set of ?predefined? features, e.g. edges (zero-crossings) or bars (peaks/throughs), matches are sought by comparing the overall configuration of contrast within delimited regions of the two images. The main disambiguating power of the model is provided by combining the results of the matchings from a number of independent channels of different coarseness (in regard to the resolution of the contrast information). The idea is that the information in the coarser channels can be used to restrict the domain of potential matches, to be considered, within the finer channels. Important for this assumption is the concept of figural continuity. To further reduce the set of potential matches, the model relies on the constraint of uniqueness. A computer implementation of the model is presented, which from the input consisting of a stereogram, produces a representation of the binocular disparity present within the stereogram. A number of results obtained from this computer implementation are also presented and discussed.