Summary form only given. I first introduce a new modeling technique which allows a total control of the volume (or mass) of objects. I also present former work in the domain. The technique is aimed at producing algebraic operators on objects in a similar way to CGS composition. In our case, I add volume control. I first study the "sum" of two objects which is such that the volume of the sum is equal to the sum of the volumes, for the "star" case. Then I give its generalisation to the case of all types of 3D objects. Then I describe how to construct the "difference" operator. An application of this concept to 3D object morphing is presented.

In this paper, we present a new method to recover an approximation of the bidirectional reflectance distribution function (BRDF) of the surfaces present in a real scene. This is done from a single photograph and a 3D geometric model of the scene. The result is a full model of the reflectance properties of all surfaces, which can be rendered under novel illumination conditions with, for example, viewpoint modification and the addition of new synthetic objects. Our technique produces a reflectance model using a small number of parameters. These parameters nevertheless approximate the BRDF and allow the recovery of the photometric properties of diffuse, specular, isotropic or anisotropic textured objects. The input data are a geometric model of the scene including the light source positions and the camera properties, and a single image captured using this camera. Our algorithm generates a new synthetic image using classic rendering techniques, and a lambertian hypothesis about the reflectance model of the surfaces. Then, it iteratively compares the original image to the new one, and chooses a more complex reflectance model if the difference between the two images is greater than a user-defined threshold.

The image segmentation problem may be considered as the search for a way to subdivide an image domain into regions which represent the projection of visible parts of objects in a real scene. The authors analyze the problem of image segmentation in the framework of the approximation theory as defined by D. Mumford and J. Shah (1988). They show that for real images the problem of the choice of the energy functional is dictated by the model of the world, and they propose a method to optimize it based on a deterministic algorithm processed at multiple levels of resolution. Problems encountered in dealing with real scenes lead to several modifications of the algorithm and the energy functional. Images are shown on which the algorithm was tested.>

The authors present region-based image processing algorithms participating in task sequencing for stereo vision. The algorithms described are prior to stereo matching. Requirements in image similarity provide helpful additional knowledge for their improvement. A recursive region division algorithm using a threshold method based on contrast maximization is described. The regions are processed in parallel and absorb their noise before being thresholded. Subsequent morphological processes improve similarity and matching results. Additional knowledge is produced by analytical processes, and is useful for both segmentation and match control. Results are presented for a stereo pair of gray-level images.>

The scope concerns both applications related to computer vision such as vision for the robots of the next century and computer graphics such as audio-visuals or simulators of physical phenomena. It is a large field of new applications where refined 3-D recognition and interpretation (not yet generally solved today) will be required. The problem that the author solves more precisely is to understand and recognize a 3-D scene observed by two cameras in a stereo position, using a given model of the scene.>

Stereo matching is the process of determining correspondences between entities in related images. Often, this is treated as two quite independent subprocesses, namely, segmentation, followed by matching. These processes are treated as naturally related, in that partial matching is fed back to the segmentation and both proceed simultaneously in a cooperative fashion. Regions are considered as the primitives to be matched, since many of the shortcomings inherent in approaches based on points or lines can be overcome by taking more developed entities. The implementation is based on maintaining a hierarchy of segmented regions in each image, corresponding to analysis at differing scales. The selection of a particular segmentation in each image at a scale appropriate to each region is validated with reference to the optimal matching region in the other image. Examples of these methods, applied to a synthetic image (incorporating color) and to natural office scenes, are presented.>

This paper presents a new method for synthesizing hierarchical textures on 3-D surfaces. This method utilizes both a mapping technique for rendering the macroscopic structure on the surface and a generalization of the direct 3-D microscopic synthesis algorithms (presented in earlier publications) for generating a homogeneous texture inside each pattern. To produce the macroscopic structure on the 3-D shape a new mapping technique is proposed. With this technique patterns distortion is minimized locally. Finally a solution to the aliasing problem adapted to our mapping is given.

We have previously shown the validity of a statistical model proposed by us for realistic textures. We have also shown how to use this model to map realistic textures onto a still surface of any shape and extent, maintaining consistency and scaling without producing repeating patterns.The problem we have to solve in this paper is how to use this statistical model in order to produce animated objects while keeping temporal consistency. This problem a priori difficult to solve as our mapping technique is statistical.We propose two different techniques. The first method called the image space method creates texture directly on the image planes. The second one called the object space method performs the texture synthesis on the object surface itself which has to be sampled finelly beforehand.

We present some new techniques for shape approximation with fractals, using Iterated Function System, a powerful method which allows good control on the resulting fractal. The main point discussed here can be stated as follows : given a grey level image A, find a few number of functions and associated probabilities that approximately generate A. Two directions have been explored : the first uses a gradient method, thus it was necessary to define a smooth error function ; the second one is based upon the ideas of simulated annealing. We then generalize the methods to a broader class of functions, and present some results.

The technique presented solves the problem of texture segmentation in two steps. In the first, a textured image is divided into small squares (20*20 in this case) and a hierarchical clustering algorithm related to a choice of ultrametric distances is used to obtain an initial segmentation. In the second step, the texture boundaries are improved using a recursive procedure with vertical and horizontal segments of smaller and smaller size, converging towards the correct texture boundaries.>

The use of fractals to synthesize complex objects is of current interest in the computer graphics community. A powerful way to compute fractals is the use of IFS (iterated function system) which is a set of contractions with associated probabilities which characterize the fractal. This theory, developed by M. Barnsley and al., can produce very complicated objects. We present a method to solve the inverse problem for these globally constructed fractals : given a set A (attractor), find an IFS that will approximately generate A. We use an optimisation method to minimize a distance between A and the current set L. Several distances have been tested and an algorithm has been implemented which gives good results. A test image is presented.

This paper presents a new method for the synthesis of textures on 3-D surfaces. With the exception of fractals, graftals and particle systems used for very restricted types of textures, one basic technique for the faithful reproduction of textures has been presented in the literature up to now. In this standard method, textures are synthesized by mapping a rectangular template onto a patch of the curved surface. In the method proposed here, the synthesis is achieved continuously on the surface by interpolating only statistics of the texture which have a much narrower spectrum than the texture itself so that there are no aliasing effects. A vast class of textures can be reproduced with a visual appearance similar to the reference texture used. The method applies for any 3-D surface and is relatively quick and easy.

We present a general method to synthesize a stochastic process from its a priori given second-order statistics. Some new results, including a new conjecture dealing with visual discrimination of stochastic texture fields, are also given.

Earlier experiments by Julesz et al. on visual discrimination are reviewed and extended in the context of texture analysis of natural images. Stochastic methods are developed for the generation of pairs of synthetic texture fields possessing multiple gray levels, significant spatial correlation, and joint moments or probability densities of controllable form. Results of several visual discrimination experiments with these computer generated texture fields are then presented. It is demonstrated, for the stochastic models investigated, that humans cannot effortlessly discriminate between pairs of spatially correlated texture fields with differing third-order probability densities when their lower order densities are pairwise equal. This is a further verification of Julesz's conjecture. Also, it is shown that human observers are sensitive to relatively small changes in the spatial autocorrelation function. Examples are presented of discriminable texture fields with identical means, variances, autocorrelation functions, and third-order nearest neighbor moments.