Technical Reports

  Gilson Antonio Giraldi

1. Scientific Visualization and Fluid Animation

·  1.1 Data Integration Middleware System for Scientific Visualization (2005)
Authors: Gilson A. Giraldi, Fabio Porto, Bruno Schulze, Vinicius
Fontes, Marcio L. Dutra

Abstract: In this paper we focus on distributed scientific visualization on the
grid. Specifically, we firstly find out basic
requirements for distributing graphics applications over a grid environment.
Then, we propose a middleware infrastructure adapted for supporting
scientific visualization applications that meets these requirements. We
claim that we should consider scientific visualization in grid from an
integrated global view of data and programs published by heterogeneous and
distributed data sources. This idea can be implemented by CoDIMS which is an
environment for the generation of Configurable Data integration Middleware
Systems. CoDIMS adaptive architecture is based on the integration of special
components managed by a control module that executes users workflows. We
exemplify our proposal with the CoDIMS-G, which is a middleware that follows
CoDIMS architecture and was designed to provide data and program integration
service for the grid. An specific configuration of CoDIMS-G for distributed
particle tracing within a grid environment is also presented.


·  1.2 Computational Animation of Fluids through Smoothed Particle Hydrodynamics (2005) - In Portuguese

Abstract: This work presents an implementation of a Computational Fluid
Dynamics (CFD) method for fluid animation. This method is based on
the Smoothed Particle Hydrodynamics (SPH) technique. The fluid
behavior is modeled through Navier-Stokes equations under
some initial conditions and constrains. The discretization is performed through SPH
The result is a representation through a particles system which move under the influence of
forces, such as gravity and pressure. If combined with efficient techniques for fluid
visualization, this methodology for animation may achieve a high
degree of realism, since it is based on physical principles for flow
representation. Preliminary results obtained reproduced from the literature show
that this is a promising technique.



Resumo: Este texto tem como tema central a animação de fluidos via  técnicas para visualização de
dados científicos (Visualização Científica) e métodos  em Mecânica Computacional. Por “animação de  fluidos”
entendemos  a geração de uma seqüência de imagens digitais contendo fluidos em movimento. Este movimento
deve ser convincente, ou seja, o fluido deve escoar com o grau de realismo necessário para o contexto do filme
que está sendo gerado. Dentro desta perspectiva, a animação de fluidos tem natureza multidisciplinar, e seu
desenvolvimento depende da interação entre profissionais das áreas de computação e engenharia. Objetivamente,
para atingir o grau de realismo necessário, é preciso que as equações de fluidos sejam convenientemente tratadas e
que as técnicas de visualização de dados utilizadas possam gerar (renderizar) a cena final com a aparência  necessária.
Estas técnicas de visualização são oriundas da computação gráfica. As equações  de fluidos (Navier-Stokes) são derivadas
de conceitos da mecânica clássica. Neste sentido temos como objetivos a apresentação de modelos matemáticos de fluidos do ponto
de vista da computação gráfica e a descrição de técnicas de visualização focadas para rendering de cenas envolvendo fluidos.



·  Um Modelo de Simulação para Escoamento Superficial de Águas sobre Terrenos Baseado em GPU
Authors: Bruno Barcellos S. Coutinho, Gilson Antônio Giraldi, Antônio L. Apolinário

Jr., Paulo Sérgio S. Rodrigues.

Abstract: In the last years a significant number of models to produce a realistic animation of fluids was proposed. In general, such modelsare based on partial differential equations and numerical methods. Recently, we proposed a discrete model based on a particle system, to simulate the superficial fluid flow over a terrain. A Digital Terrain Model or DTM is used as the basis of the simulation process. The fluid dynamics is ruled by a particle system based on a cellular automata. The particles runs through the edges of lattice, depending on terrain declivity and the behavior of the fluid in the particle'sneighborhood. In this paper, we present a GPU-based implementation of that model. The terrain geometrical and physical properties are structured in a way that the whole simulation process can be done in the GPU. The simplicity of the automata model allied to the computational power of GPU allow us to generate real-time animations, as the  experimental results demonstrate.


2. Deformable Models


Authors: Gilson A. Giraldi and Antonio Alberto F. Oliveira

Abstract: This paper presents a convexity analysis for the dynamic snake model based on the Potential Energy functional and the Hamiltonian formulation of the classical mechanics. First we see the snake model as a dynamical system whose singular points are the borders we seek. Next we show that a necessary condition for a singular point to be an attractor is that the energy functional is strictly convex in a neighborhood of it, that means, if the singular point is a local minimum of the potential energy. As a consequence of this analysis, a local expression relating the dynamic parameters and the rate of convergence arises. Such results link the convexity analysis of the potential energy and the dynamic snake model and point forward to the necessity of a physical quantity whose convexity analysis is related to the dynamic and which incorporate the velocity space. Such a quantity is exactly the (conservative) Hamiltonian of the system.

·  2.2 Dual and Topologically Adaptable Snakes and Initialization of Deformable Models (2000)
Authors: G. A. Giraldi and N. Vasconcelos and E.Strauss and A. F. Oliveira

Abstract: The original proposal of Active Contour Models, also called snakes, suffers from the strong sensitivity to the initial contour position
and can not deal with topological changes. In this paper we describe some techniques to address these limitations. The problem of topological changes is addressed by the T-Snakes model by embedding the snake in the framework of a simplicial decomposition of the domain. The sensitivity to initialization is addressed by the Dual method by using two contours going towards the desired boundary. We integrated these two methods, given a new approach called Dual-T-Snakes, which maintains the topological capabilities of the T-Snakes and the ability of avoiding local minima of the Dual Active Contour model. One advantage of the Dual-T-Snakes is the reduction of the search space, which allows a more efficient application of a dynamic programing technique. To initialize the Dual-T-Snakes automatically we developed a methodology based on a multiresolution scheme and on the T-Snakes framework. Just as the T-Snakes, our methods (Dual-T-Snakes and initialization framework) can be extended to 3D. We demonstrate our methodologies in synthetic and medical images showing also how to incorporate fuzzy and multi-scale techniques in the Dual-T-Snakes approach. Finally, we discuss some challenges in active contour models and present the conclusions of our work.

·  2.3 Mathematical Elements of Deformable Models for Image Processing (2009)

Authors: G. A. Giraldi and Paulo S. S. Rodrigues

Abstract: Active Contour Models, also called Snake models, are powerful techniques for boundary extraction and segmentation of 2D images. Despite of their abilities, the non-invariance of the internal energy under affine transformations, the non-convexity of the model energy functional and the inability to deal with topological changes are known limitations for most of these methods. In this book we describe some techniques to address these limitations. The non-invariance of the internal energy has been addressed in the context of active shape models and Lie groups. The non-convexity of the model energy can be addressed through Dual contour approaches, diffusion approaches, as well as an automatic procedure to initialize the model closer to the desired boundary. The problem of topological changes is addressed by embedding the snake in the framework of a simplicial decomposition of the domain or through implicit formulations. In this text, we offer some background in parametric snakes, followed by a taxonomy of deformable models for image processing. Then, we discuss mathematical elements behind the main works that address the mentioned problems. Next, we survey snake approaches according to the presented taxonomy. In order to complete the material, we dedicate a Chapter for extensions to Deformable Surface approaches. Finally, this lecture ends with a discussion on snake approaches, some drawbacks and future works. We survey applications for shape recovery in cell and ultrasound images.


3. Image Representation and Analysis

3.1 Fourier Analysis for Linear Filtering and Holographic Representations of 1D and 2D Signals

Abstract: In this paper, we focus on Fourier analysis and holographic transforms for
signal representation. For instance, in the case of image processing, the holographic representation
has the property that an arbitrary portion of the transformed image enables
reconstruction of the whole image with details missing. We focus on holographic representation
de[1]ned through the Fourier Transforms. Thus, we firstly consider the Fourier
analysis in the context of signal processing. Next, we review the Discrete Holographic
Fourier Transform (DHFT) for image representation. Then, we describe the contributions
of our work. We show a simple scheme for progressive transmission based on the DHFT.
Next, we propose the Continuous Holographic Fourier Transform (CHFT) and discuss
some theoretical aspects of it for 1D signals.
Finally, some testes are presented in the
experimental results.

3.2 Aprendizado de Variedades: Aspectos Geométricos e Aplicações

Resumo: Muitas áreas, tais como visão computacional, reconhecimento de padrões e análise de

imagens, requerem manipulação de uma grande quantidade de dados representados originalmente

em espaços de dimensão elevada. Em particular, para um banco de imagens de faces

humanas, existe um considerável nível de redundância oriundo da própria anatomia (boca,

nariz, orelhas, olhos, etc). Neste sentido, a redução de dimensionalidade é extremamente

necessária para possibilitar a análise dos dados. Supondo que os dados possam ser ajustados

a uma variedade de dimensão d contida num espaço Euclideano de dimensão muito maior que d,

podemos estimar o espaço tangente local e definir uma topologia sobre a variedade. Neste

contexto, as técnicas de Aprendizado de Variedades (Manifold Learning) surgem como um

recurso importante que envolve áreas como aprendizagem de máquina e mineração de dados

para tratar a redução de dimensionalidade recuperando a geometria intrínseca do conjunto de

dados. Exemplos clássicos de técnicas de aprendizado de variedades são o ISOMAP e LLE.

Ilustramos uma aplicação à imagens de faces humanas, onde geometria de aquisição permite

estimar propriedades da variedade que representa os dados. Uma das técnicas recentes de

aprendizado de variedades é o RML, a qual visa preservar a geometria intrínseca usando

conceitos clássicos de Geometria Riemanniana. Esta técnica produz bons resultados e será

apresentada neste trabalho. Apresentaremos ainda uma fundamentação teórica para aprendizado

de variedades e redução de dimensionalidade, seguida de uma revisão das principais técnicas na área.

Serão também listados alguns dos principais desafios neste tema.


4. Quantum Neural Networks


·  3.1 Quantum Models for Artificial Neural Networks (2002)
Authors: Gilson A. Giraldi and Jean Faber

Abstract: There has been a growing interest in artificial neural networks
(ANNs) based on quantum theoretical concepts and techniques due to cognitive
science and computer science aspects. The so called Quantum Neural Networks
(QNNs) is an exciting area of research in the field of quantum computation
and quantum information. However, a key question about QNNs is what such an
architecture will look like as an implementation on quantum hardware. To
look for an answer to this question we firstly review some basic concepts in
ANNs and emphasize their inherent non-linearity. Next, we analyze the main
algorithms and architecture proposed in this field. The main conclusion is
that, up to now, there is no a complete solution for the implementation of
QNNs. We found partial solution in models that deal with nonlinear effects
in quantum computation. The Dissipative Gate (D-Gate) and the Quantum Dot
Neural Network are the focused models in this field. The former is a
theoretical one while the later is a device composed by a quantum dot
molecule coupled to its environment and subject to a time-varying external
field. A discretized version of the Feynman path integral formulation for
this system can be put into a form that resembles a classical neural
network. Starting from these models, we discuss learning rules
in the context of QNNs. Besides, we present our proposals in
the field of QNNs.