Accession Number : ADP001194

Title :   Computing Visual Correspondence,

Corporate Author : MASSACHUSETTS INST OF TECH CAMBRIDGE ARTIFICIAL INTELLIGENCE LAB

Personal Author(s) : Kass,Michael

Report Date : JUN 1983

Pagination or Media Count : 7

Abstract : A computational framework for solving the visual correspondence problem is presented and evaluated by using a stochastic image model. The framework differs from previous work in that it emphasizes the combination of a large collection of independent measurements. Partial derivatives of images smoothed with a few different-sized Gaussian filters are suggested as suitable measurements. A specific computation is shown based on a stochastic image model to reliably establish whether or not two points correspond, provided that the signal to correspondence noise ratio in the images to be matched exceeds two. The computation has been applied to artificial and natural images with encouraging results. (Author)

Descriptors :   *Mathematical models, *Stochastic processes, *Images, *Visual perception, Computations, Algorithms, Problem solving, Errors, Vision, Light, Intensity, Statistical processes, Aerial photographs

Distribution Statement : APPROVED FOR PUBLIC RELEASE