A computer vision system has been implemented that can recognize threedimensional objects from unknown viewpoints in single gray-scale images. Unlike most other approaches, the recognition is accomplished without any attempt to reconstruct depth information bottom-up from the visual input. Instead, three other mechanisms are used that can bridge the gap between the two-dimensional image and knowledge of three-dimensional objects. First, a process of perceptual organization is used to form groupings and structures in the image that are likely to be invariant over a wide range of viewpoints. Second, a probabilistic ranking method is used to reduce the size of the search space during model based matching. Finally, a process of spatial correspondence brings the projections of three-dimensional models into direct correspondence with the image by solving for unknown viewpoint and model parameters. A high level of robustness in the presence of occlusion and missing data can be achieved through full application of a viewpoint consistency constraint. It is argued that similar mechanisms and constraints form the basis for recognition in human vision. This paper has been published in Artificial Intelligence, 31, 3 (March 1987), pp. 355–395. 1 1

To develop more satisfying image navigation systems, we need tools to construct a semantic bridge between the user and the database. In this paper we present an image indexing scheme and a query language, which allow the user to introduce a cognitive dimension to the search. At an abstract level, this approach consists of: 1) learning the "natural language" that humans speak to communicate their semantic experience of images, 2) understand the relationships between this language and objective measurable image attributes, and then 3) develop the corresponding feature extraction schemes. In our previous work we have conducted a number of subjective experiments in which we asked human subjects to group images, and then explain verbally why they did so [1]. The results of this study indicated that part of the abstraction involved in image interpretation is often driven by semantic categories, which can be broken into more tangible semantic entities, i.e. objective semantic indicators. By analyzing our experimental data, we identified some candidate semantic categories (i.e. portraits, people, crowds, cityscapes, landscapes, etc.), discovered their underlying semantic indicators (i.e. skin, sky, water, object, etc.), and derived important lowlevel image descriptors accounting for our perception of these indicators. In our recent work we have used these findings to develop a set of image features that match the way humans communicate image meaning, and a "semantic-friendly" query language for browsing and searching diverse collections of images. We have implemented our approach into an Internet search engine, ISee, and tested it on a large number of images. The results we obtained are very promising.

This paper surveys the techniques which have been applied to the problem of recognising three-dimensional objects in two-dimensional images. Human vision was discussed in the works of the ancient Greek philosophers, and has also been of interest to modern philosophers. The Gestalt school of psychology in the early part of the twentieth century provided a number of useful insights into human perception. Computer vision research effectively started with the pioneering work of Roberts, who built a program capable of recognising simple objects in a blocks world. The blocks world paradigm provides a simplified model in which new approaches can be tested, and has been adopted from time to time by a number of researchers. The dominant paradigm in modern computer vision research is that pioneered by Marr, and known as inverse optics, or the Marr paradigm. In this approach, edges, surfaces and depth cues are identified before object recognition is attempted. Central problems in much of this wor...

This paper describes a new method of perceptual organization applied to the extraction of thin networks on aerial and medical images. The key point of our approach is to consider perceptual grouping as a problem of optimization. First the quality of a grouping is defined with a class of functions inspired by the energy functions used for active contours optimization (involving curvature, co-circularity, grey levels, and orientation). Such functions can be computed recursively, and optimized from a local to a global level with an algorithm related to dynamic programming. This is followed by a selection procedure which rates and extracts principal groupings. The validity of our approach is presented with synthetic images, aerial and medical data. Keywords: Perceptual Organization, Dynamic Programming, Non convex Objects Extraction, Segmentation. 1 Introduction The detection of thin networks is important for many problems of image analysis. For this class of problems, the lines detected...

We present a method of grouping edge elements upon the relation of smoothed continuity. This method is based upon the optimization of a quality function of curvature and edge intensity, from a local to a global level. It is implemented with a network of locally-connected processing elements. This method was first designed to deal with dot images and then was applied to the problem of road extraction in satellite images. The extension to satellite images has been achieved by means of constraints from low level vision algorithms and a dynamic exploitation of results of the grouping process. This method copes with a complex combinatorial problem allowing an efficient and parallel solution to be found. Experimental results on synthesis and natural satellite images show the validity of the approach and its possible incorporation into general computer vision systems. 1 Introduction Since 1923, psycho-visual experiments [13] have shown that human perception is driven by grouping phenomena of...

This paper describes a new method of perceptual organization of thin networks using geometric properties. The key point of our approach is to consider perceptual organization as a problem of optimization : solutions to this problem are the best matchings between continuous curves and the low level primitives. First the quality of a grouping is defined with a class of functions related to the energy functions of active contours optimization. Such functions are computed recursively, and optimized from a local to a global level with an algorithm related to dynamic programming. This is followed by a selection procedure which rates and extracts principal groupings automatically and gives a new segmentation of image primitives, based on smoothed continuation. This segmentation is used to initialize a high level interpretation process involving projective reconstruction of 3D contours in sequences of images. The adaptability and robustness of this method have been tested on various situations...

The detection and the identification of an object in an image is an extremely complex combinatory problem. The objective of a shape recognition system is to reduce this complexity and make the problem solvable. In this paper, we present a two stage method for features extraction. First we start with a description of the scene using Perceptual Organization techniques. This step extracts a set of salient structures in the image, from a local to gobal level, without prior knowledge of the scene. We show how this description is more appropriate for the shape extraction level. Along the analysis, increasingly complex features are extracted from this description, organized into a graph of structural relationships, and validated using the original image. The final description of objects in the image can be used for matching a 3D model with a 2D scene or used to track objects in a sequence. Keywords: Perceptual Grouping, Shape extraction, Dynamic Programming, Curves Detection. 1 Introduction ...

We present a new method for Perceptual Grouping of pixels into roads after crest lines detection in satellite pictures. First the visual properties expected from the groupings are modelled as a quality function similar to active contour functions. They involve curvature, grey levels and co-circularity. This function is computed recursively and optimized from a local to global level with an algorithm related to dynamic programming. The final groupings are then selected according to their global quality. Applied to satellite images, the method proved its adaptability and its robustness to noisy environments. The results showed how the use of visual properties can provide an effective segmentation with no prior knowledge of the scene. This segmentation can be used to initialize a high level interpretation process or give a first description of the scene to an interactive decision system. Keywords: Perceptual Organization, Dynamic Programming, Roads Extraction, Segmentation, Remote Sens...