Sticks, Plates, and Blobs: A Three-Dimensional Object Representation for Scene Analysis Linda G. Shapiro Prasanna G. Mulgaonkar John D. Moriarty Robert M. Haralick Virginia Polytechnic Institute and State University Department of Computer Science how the parts fit together. Our models have three ABSTRACT kinds of three-dimensional Parts: sticks, Plates. In this paper, we describe a relational modeling technique which categorizes three-dimensional objects at a gross level. These models may then be used to classify and recognize two dimensional views of the object, in a scene analysis system. I. Introduction The recognition of three-dimensional objects from two-dimensional views is an important and still largely unsolved problem in scene analysis.
INTERPRETIVE VISION AND RESTRICTION GRAPHS Rodney A. Brooks and Thomas 0. Binford Artificial Intelligence Laboratory, Computer Science Department Stanford University, Stanford, California 94305 ABSTRACT We describe an approach to image interpretation which uses a dynamically determined interaction of prediction and observation. We provide a representational mechansim, built on our geometric modeling scheme which facilitates the computational processes necessary for image interpretation. The mechanism implements generic object classes and specializations of models, enables case analysis in reasoning about incompletely specified situations, and manages multiple hypothesized instantiations of modeled objects in a single image. It is based on restriction nodes and quantified variables. A natural partial order on restriction nodes can be defined by comparing the satisifiability of their constraints. Nodes are arranged in an incomplete restriction graph whose arcs represent relations of nodes under the partial order. Predictions are matched to descriptions by finding maximal isomorphic subgraphs of a prediction graph and an observation graph 183 subject to a naturally associated infimum of restriction nodes being satisifiable. In this manner constraints implied by local two dimensional matches of image features to predicted features are propagated back to the three dimensional model enforcing global consistency. A. Image Interpretation I INTRODUCTION A descriptive process is one which takes an image and produces a description of image features and their relations found within that image. A predictive process is one which uses models of objects expected in an image to predict what features and their relations will be present in the image.
Ullman  has investigated the shape of subjective contours (see for example . In fact, the work is more generally applicable to other cases of pcrccptual shape completion, in which the visual system is not constrained by actual physical intensity changes. Examples include patterns foimcd from dots and incomplctcly line drawings and alphabetical characters. The foim of the solution derives from a number of premises, one of which Ullman calls "the locality hypothesis". 'I'hc"cxpcrimcntal observation" rcfcrrcd to is the following: suppose that A' is a point near A on the filled-in contour AB as shown in Figure 1.
WHAT SHOULD BE COMPUTED IN LOW LEVEL VISION SYSTEMS William B. Thompson Albert Yonas ABSTRACT Recently, there has been a trend towards developing low level vision models based on an analysis of the mapping of a three dimensional scene into a two dimensional image. Emphasis has been placed on recovering precise metric spatial information about the scene. While we agree with this approach, we suggest that more attention be paid to what should be computed. Pschophysical scaling, adaptation, and direct determination of higher order relations may be as useful in the perception of spatial layout as in other perceptual domains. When applied to computer vision systems, such processes may reduce dependance on overly specific scene constraints.
Certain image properties, such as parallelisms, symmetries, and repeated patterns, provide cues for perceiving the 3-D shape from a 2-D picture. This paper demonstrates how we can map those image properties into 3-D shape constraints by associating appropriate assumptions with them and by using appropriate computational and representational tools. We begin with the exploration of how one specific image property, "skewed symmetry", can be defined and formulated to serve as a cue to the determination of surface orientations. Then we will discuss the issue from two new, broader viewpoints. One is the class of Affine-transformable patterns. It has various interesting properties, and includes skewed symmetry as a special case. The other is the computational paradigm of shape-from-texture. Skewed symmetry is derived in a second, independent way, as an instance of the application of the paradigm. This paper further claims that the ideas and techniques presented here are applicable to many other properties, under the general framework of the shape-from-texture paradigm, with the underlying meta-heuristic of non-accidental image properties.
A STATISTICAL TECHNIQUE FOR RECOVERING SURFACE ORIENTATION FROM TEXTURE IN NATURAL IMAGERY Andrew P. Witkin Artificial Intelligence Center SRI International, Menlo Park, CA 94025 ABSTRACT A statistical method is reported for inferring the shape and orientation of irregularly marked surfaces using image geometry. The basis for solving this problem lies in an understanding of projective geometry, coupled with simple statistical models of the contour generating process. This approach is first applied to the special case of surfaces known to be planar. The distortion of contour shape imposed by projection is treated as a signal to be estimated, and variations of non-projective origin are treated as noise. The resulting method is next extended to the estimation of curved surfaces, and applied successfully to natural images.