This paper describes the underlying ideas and algorithmic details of a computer program that performs at a human level of competence for a significant subset of the curve partitioning task. It extends and rounds out the technique and philosophical approach originally presented by Fischler and Bolles (1986). In particular, it provides a unified strategy for selecting and dealing with interactions between salient points, even when these points are salient at different scales of resolution. Experimental results are presented involving on the order of 1000 real and synthetically generated images.< >

In this paper we address a basic problem in machine perception: the tracing of perceptually obvious "line-like" structure appearing in an image; we present some new ways of looking at this problem and provide techniques that are significantly more general and effective than previously reported methods for this task. Our approach is a departure from the procedures usually employed in the following important respects: (1) We recognize the typically overlooked distinction between lines and edges; finding lines offers significant simplifications not available when searching for edges. (2) The perceptual primitive we extract from the image to indicate the local presence of linear structure is a function of total intensity variation over an extended portion of the image rather than intensity gradient information which can be extracted with a mask-like "local image operator." While easy to compute, this new primitive appears to be more effective and stable than the almost universally employed gradient based primitives. (3) The overall procedure can be applied to an arbitrary image with essentially no human intervention (parameter tuning or attention focusing) and produces excellent results. Examples of the delineation procedure are shown for aerial, industrial, and radiographic imagery.

Our approach is a departure from the procedures usually employed in the following important respects: (1) We recognize the typ ica l ly overlooked d is t inc t ion between lines and edges; f inding l ines offers s igni f icant simpl i f icat ions not available when searching for edges. (2) The perceptual pr imit ive we extract from the image to Indicate the local presence of l inear structure is a function of to ta l intensi ty var iat ion over an extended portion of the image rather than Intensity gradient Information which can be extracted with a mask-like " local image operator." While easy to compute, th is new pr imit ive appears to be more effect ive and stable than the almost universal ly employed gradient based pr imit ives. (3) The overal l procedure can be applied to an arb i t rary image with essential ly no human intervention (parameter tuning or attent ion focusing) and produces excellent resul ts . Examples of the delineation procedure are shown for ae r ia l , i ndus t r ia l , and radiographic imagery.

Aerial and satellite imagery provide an economical means of gathering large amounts of data on the earth's resources and environment. However, except in the area of survey tasks such as crop inventories and land use that can be performed with multispectral analysis, there are few economically feasible techniques for automatically extracting the useful information from such imagery.

New techniques are described for some of the important computations in the automatic determination of image-to-database correspondences. In particular, a technique to predict a region in the image within which a feature is expected to appear, a set of techniques to verify feature matches, and a technique to extend the refinement process to include a new type of match based on linear features, such as roads, are discussed. These techniques are demonstrated in an example in which the system reduces the uncertainties from approximately plus or minus 200 feet on the ground to approximately plus or minus two feet.

Parametric correspondence is a technique for matching images to a three dimensional symbolic reference map. An analytic camera model is used to predict the location and appearance of landmarks in the image, generating a projection for an assumed viewpoint. Correspondence is achieved by adjusting the parameters of the camera model until the appearances of the landmarks optimally match a symbolic description extracted from the image. The matching of image and map features is performed rapidly by a new technique, called chamfer matching, that compares the shapes of two collections of shape fragments, at a cost proportional to linear dimension, rather than area. These two techniques permit the matching of spatially extensive features on the basis of shape, which reduces the risk of ambiguous matches and the dependence on viewing conditions inherent in conventional image based correlation matching.