This paper presents a novel approach to supervised learning of image-understanding tactics based on the use of contextual information. We use a database to store past experiences. From this database and the context elements computed from the task specifications and the input data, we determine whether or not an algorithm is applicable, and which parameters are suitable for it. The database is continuously updated with information of success or failure of the system.

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.>

Results from an ongoing project concerned with recognizing objects in complex scene domains, especially in the domain that includes the natural outdoor world, are described. Traditional machine recognition paradigms assume either that all objects of interest are definable by a relatively small number of explicit shape models or that all objects of interest have characteristic, locally measurable features. The failure of both assumptions has a dramatic impact on the form of an acceptable architecture for an object recognition system. In this work, the use of the contextual information is a central issue, and a system is explicitly designed to identify and use context as an integral part of recognition that eliminates the traditional dependence on stored geometric models and universal image partitioning algorithms. This paradigm combines the results of many simple procedures that analyze monochrome, color, stereo, or 3D range images. Interpreting the results along with relevant contextual knowledge makes it possible to achieve a reliable recognition result, even when using imperfect visual procedures. Initial experimentation with the system on ground-level outdoor imagery has demonstrated competence beyond what is attainable with other vision systems.>

In this paper we offer a critical evaluation of the partitioning (perceptual organization) problem, noting the extent to which it has distinct formulations and parameterizations. We show that most partitioning techniques can be characterized as variations of four distinct paradigms, and argue that any effective technique must satisfy two general principles. We give concrete substance to our general discussion by introducing new partitioning techniques for planar geometric curves, and present experimental results demonstrating their effectiveness.

A single two-dimensional image is an ambiguous representation of the three-dimensional world¿many different scenes could have produced the same image¿yet the human visual system is ex-tremely successful at recovering a qualitatively correct depth model from this type of representation. Workers in the field of computational vision have devised a number of distinct schemes that attempt to emulate this human capability; these schemes are collectively known as ``shape from...'' methods (e.g., shape from shading, shape from texture, or shape from contour). In this paper we contend that the distinct assumptions made in each of these schemes is tantamount to providing a second (virtual) image of the original scene, and that each of these approaches can be translated into a conventional stereo formalism. In particular, we show that it is frequently possible to structure the problem as one of recovering depth from a stereo pair consisting of the supplied perspective image (the original image) and an hypothesized orthographic image (the virtual image). We present a new algorithm of the form required to accomplish this type of stereo reconstruction task.

This paper presents a fault tolerant multiprocessor architecture suitable for real time control applications requiring an extremely high degree of reliability. The architecture satisfies the following requirements:

The primary problem dealt with in this paper is the following. Given some description of a visual object, find that object in an actual photograph. Part of the solution to this problem is the specification of a descriptive scheme, and a metric on which to base the decision of "goodness" of matching or detection.

Abstract—This note establishes the fact that an optimal synthesis technique (even for a small number of variables) must abandon the basic assumption of assigning only +1 weights to the primary inputs. Examples are presented where both negative primary weights and/or nonunitary primary weights are required to achieve minimality. A method of network analysis is presented which leads to optimal synthesis in a significant number of new cases. The previously accepted lower bound on the number of devices needed to realize a symmetric function is questioned, and a proof is given to establish its validity. A previously published assertion about a least upper bound on the number of devices is shown to be incorrect.

This paper treats the problem of compressing binary strings of fixed length into shorter strings of fixed length. Constraints are assumed with respect to the maximum number of ones which can occur in the input binary string and on the memory available for encoding or decoding using table look-up operations. Bounds and coding techniques are presented. There appear to be many applications to the information sciences and especially to memory utilization in digital computers and to information retrieval.

This paper discusses a class of pattern-recognition decision-making procedures which can be efficiently mechanized by a network of linear threshold devices. A theoretical discussion of four decision techniques (which are members of the above-mentione class) is augmented by an experimental program which investigates questions not amenable to analysis.

In this paper, the pattern recognition problem is considered to be composed of two subproblems. The first subproblem is one of abstracting significant features or characteristics from the patterns being dealt with, while the second subproblem is concerned with identifying the pattern which gave rise to a particular set of features (i. e., a decision-making problem). The decision-making system to be discussed is a special purpose digital computer, which simulates a network of threshold elements designed to correctly identify a preselected set of "typical" patterns. In addition to describing the design procedures, methods of feature extraction, and the hardware employed, the resuits of experiments involving several thousand handwritten characters will also be presented.