Special Session on Computer Graphics


July, 26 - Monday

16:30 - 17:00
Waldemar Celes , TeCGraf/PUC-Rio
Distributed Visualization

Nowadays, a large number of applications require multiple-display visualization system capable of providing immersion sensation. The common approach to provide multiple synchronized views uses a powerful centralized processing unit to support the rendering process on all screens. This talk focuses on an alternative approach: a distributed architecture that supports a flexible and reliable visualization system. First, we briefly introduce some issues related to virtual world modeling and discuss different ways to store the scene, its components and appearance. We then extend the discussion to model animation and its importance in a distributed environment. We also present a few concepts related to virtual world visualization such as user-observer mapping and immersion sensation. Finally, we describe the distributed approach we have been using to achieve multiple-display system that, while giving the users a sensation of immersion, uses heterogeneous network of low-end graphics workstations.

17:00 - 17:30
Marcelo Walter , Universidade do Vale do Rio dos Sinos - UNISINOS
Integrating Shape and Pattern in Mammalian Models

The giraffe and its patches, the leopard and its spots, the tiger and its stripes are spectacular examples of the integration of a visual pattern and a body shape. We will present in this talk a method for integrating a pattern generation system with a body growth and animation system. The pattern generation system can deliver a variety of animal coat patterns which are quantitatively validated and biologically plausible. The growth and animation system uses experimental growth data to produce individual bodies and their associated patterns automatically. We use the example of the giraffe to illustrate how the approach take us from a canonical embryo to a full adult giraffe in a continous way with results that are realistic looking. The method also allows a considerable amount of user control for pattern generation and body shapes.

July, 27 - Tuesday

16:30 - 17:10
Silvio de Barros Melo , Universidade Federal de Pernambuco
Piecewise Trilinear Deformation of Tomographic Models

In this work we introduce an iterative method that deforms brain models built from tomographic images. The deformation is used for normalization purposes: individual models are deformed to match the shape, orientation and internal morphology of a reference model. In this method the individual and the reference models are each enclosed in a cube which is subdivided to form a rectangular grid. The vertices in the individual models grid are perturbed and the contents of each cell is then trilinearly mapped into a cube. The composite of all resulting cubes form the deformed model to be compared with the reference. The perturbations on the vertices are generated by a simulated annealing optimization technique. To maximize the performance the models are represented in a multi-resolution fashion and the method is paralleled.

17:20 - 18:00
Geovan Tavares , Laboratorio MatMidia, PUC-Rio
Volume Modeling and Aplications

In recent years a wide class of problems in biology, medicine, geology, geophysics, among others, has demanded the handling of a huge amount of data. This kind of data, even though they have, typically, some organization, they are hard to parametrize in order to make a flow in/flow out of meaningful information. In this talk we will show how to design a mathematically sound data structure which integrate in a unique computational environment surface and volume modeling, and visualization. Issues like data acquisition, image processing, edge detection, tridimensional reconstruction and rendering will be addressed. We will also point out how those ideas are applied on solve problems in petroleum geology and the medical sciences.

July, 28 - Wednesday

16:30 - 17:10
Cicero Mota , Universidade do Amazonas
Curvature Operators in Vision and Image Processing

A frequently used method in vision and imagem processing is the use of a linear operator to transform images. Such success, in part, comes from the simplicity in computing with these transforms. In spite of the successful use of linear methods, it is well known that some visual sistems cotain cells which are insensitive to image features whose variation is punctual or unidirectional. Linear operators can not offer an adequate model of these cells, thus the use of non-linear operator becomes a necessity. Experiments reveal that most of the information carried by an image is located in the points which are extrems of the curvature. Based in those experiments, C. Zetzche, E. Barth and B. Wegmann present a detailed study of the limitations of linear operators and propose the use of the Gaussian curvature of the image surface as good non-linear operator for perceptual studies. Therefore the study of this operator and its inverse assumes a fundamental role in their proposal. Also, the authors conjecture that the set of non-zero Gaussian curvature contains the necessary information to reconstruct the image completely. We address this question in both a continuos and numerical viewpoint.

17:20 - 18:00
Dibio Leandro Borges , Universidade Federal de Goias
In search for identifying Salient Visual Features

The problem of identifying which features are the most important for Computational Vision tasks is a difficult, and yet unsolved one. Although there are differences in the way they are computed, and the constraints used, visual features such as edges and curvatures are believed to be important cues for most visual perception algorithms. One central question to this is how, and in what level of detail or scale, it is possible to identify which features retain more information, or putting it in another way are salient to the observer. In this talk we address this question by proposing that the observations could be made throughout a multiresolution framework, where we detect edges and curvatures of the observed objects in different scales, and measure from one scale to another the features with highest entropy. The salient features of edges, and of curvatures could then be identified as being the ones with the highest information content. We show and analyse results in face and outdoor images.

July, 29 - Thursday

16:30 - 17:10
Luiz Henrique de Figueiredo , Laboratorio Visgraf - IMPA / LNCC
Interval methods for ray casting implicit surfaces with affine arithmetic

We study the performance of affine arithmetic as a replacement for interval arithmetic in interval methods for ray casting implicit surfaces. Affine arithmetic is a variant of interval arithmetic designed to handle the dependency problem, and which has improved several interval algorithms in computer graphics.
See also http://www.tecgraf.puc-rio.br/~lhf/sib99/

17:20 - 18:00
Paulo Cezar Pinto Carvalho , IMPA - Instituto de Matematica Pura e Aplicada
Interactive terrain visualization using client-server architecture

Interactive terrain visualization is useful for various applications such as Geographical Information Systems, military training, city planning and tourist information systems. In most of these cases, it is natural to use a Client-Server architecture, in which the data, usually very large, is stored in one or more servers, that respond to requests made by client programs running on the users' machines. This work addresses the various techniques -- data storage schemes, network communication, data visualization and interaction control -- employed in such systems. Based on this discussion, an application to allow visualization of large terrain data over a local network was developed. This system adopts a multi-resolution representation for the terrain data. This allows interaction between user and application to be kept high, regardless of the size of the data, due to the use of algorithms which select the proper level

July, 30 - Friday

16:30 - 17:00
Romildo Jose da Silva , Universidade Federal do Ceara
Video Cut Detection Using Differential Geometry

We introduce an algorithm to detect cuts in video sequence. The algorithm is based on volumetric processing of video sequence, and uses techniques from differential geometry to classify the cut frames. The volumetric approach allow us to capture the intrinsic coherence of consecutive frames in a video shot, enabling thus a robust detection of video cuts. The algorithm is very robust with respect to the presence of noise, and avoids the detection of "false cuts" in video shot.

17:00 - 17:30
Antonio Oliveira , COPPE-UFRJ
Dual and Topologically Adaptable Snake Models

In this talk we first present an edge definition based on Riemannian geometry concepts wich can be applied for vector-valued and single-valued images. This definition can be used to present some edge detection methods which can be formulated using isotropic (or anisotropic) diffusion processes. We use that more general characterization of an edge when introducing the parametric snake models for boundary extraction. These models consist of an elastic curve (or surface) which can dynamically conform to object shapes in response to internal forces (elastic forces) and external forces (image and constraint forces). These forces are either related to a functional minimization process or obtained on basis of local information. Due to the non-convexity of the energy functional, obtaining a global extremun can require the use of techniques for rejecting local minima. Among these techniques we focus the Dual Snakes one. Next, we introduce the Topologically Adaptable Snakes (T-snakes) which have the ability of dealing with topological changes. In this framework the snakes can either split or merge. Finaly, we present some experiments using the snake models described above.

Junior Barrera , IME-USP
Automatic Design of Morphological Operators for Motion Segmentation

A problem of interest in digital video edition is the elimination of moving objects from one video and the introduction of pieces of other videos in their places. A fundamental problem to build computational tools for this purpose is the segmentation of moving objects. In this talk, we approach this problem by a new technique, based on Becheur-Meyer's paradigm, with markers detected by morphological operators designed by computational learning. The objects in the first frames of the video are marked manually and used to train the markers detector. Then, the operator designed is used to mark the objects in other frames and Beucheur-Mayer's paradigm is applied to all frames marked by the detector. Some synthetical and real world examples illustrate the application of the technique proposed. Complex sistuations as oclusion are examined.