Sibgrapi 2008 - Posters

DICOM Server Applied to Medical Image Processing
Luis Sebastián Barberis, Fabián Lucena, Juan Pablo Graffigna Vaggione, María Eugenia Gómez Marrello, and Raúl Romo (Universidad Nacional de San Juan)
(pp 1 - 2). (paper)

This work describes the design and implementation of a DICOM server that manages medical image studies for their further processing. It is based on connectivity specifications of a massive clinic standard, and has the capacity to manage studies and to run different processing algorithms, according to a decision making system by means of databases. The design provides great flexibility in terms of configuration and operations assignment, and permits compatibility with DICOM images transmission systems. The goal of the development is to apply automatic algorithms of clinical image processing carried out by research groups. By means of this system, it is possible to validate clinically new algorithms and employ them for the diagnosis and patient monitoring.

Watershed Transform based Interactive Image Segmentation Tool
Bruno Klava (USP) and Nina Hirata (USP)
(pp 3 - 4). (paper)

Watershed from markers and hierarchical watershed are approaches suitable for interactive image segmentation: in the former, the user can edit markers to control the segmentation result; in the latter, the user can select an image partition from a nested set of partitions. We propose an interactive image segmentation tool that allows transition from one approach to other and thus the combination of the strengths of both.

Marvin: A Tool for Image Processing Algorithm Development
Gabriel Archanjo, Fabio Andrijauskas, and Danilo Rosetto Muñoz
(pp 5 - 6). (paper)

This work presents, Marvin, tool for image processing algorithm development. This tool is composed by a framework and an image manipulation program that supports extensibility through plug-ins. The framework, used to develop the plug-ins, provides features for image manipulation, algorithm analysis and integration with the base application. Marvin is suitable for image processing algorithm development and prototyping and for education purpose.

Color Gradient Map-Oriented Anisotropic Diffusion Filtering
Leandro Coser (UFSC), Antonio Sobieranski (UFSC), Adiel Mittmann (UFSC), Eros Comunello (UNIVALI), and Aldo Wangenheim (UFSC)
(pp 7 - 8). (paper) (poster)

Anisotropic diffusion filtering is a well-established technique for image enhancement that smooths images without destroying edge information. However, when many filtering iterations are applied, edges gradually fade away and are ultimately smoothed by the process. We propose the adoption of a color gradient map to guide the smoothing so that clearly-defined edges are preserved even after many iterations. Preliminary experiments show good results when compared with the regular anisotropic diffusion filter.

OCR4JkanjiCards: Exploring Japanese Character Recognition
Oswaldo Bassani Neto (USP), Celso Kurashima (UFABC), and Marcelo Z. do Nascimento (UFABC)
(pp 9 - 10). (paper)

The OCR4JkanjiCards system is presented in this work, as a proposal for taking photos of Japanese characters using a smartphone and have them automatically recognized within a digital Japanese dictionary. Image processing techniques and neural network strategies are necessary, as well as the exploration of several programming tools.

Fusion of Infrared and Visible Spectra Face Recognition Methods
Giovani Chiachia (USP) and Aparecido Marana (UNESP)
(pp 11 - 12). (paper)

In general, face recognition systems are based on visible spectrum images and, in order to have good performance, they need to work in light-controlled environments. However, the performance of such systems decrease significantly when illumination changes. On other hand, Long-Wave Infrared (LWIR) face imagery is insensitive to illumination changes and gives the temperature pattern from the face to be recognized. The purpose of this work is to assess the performance of the fusion of well-known statistical visible and LWIR-based methods for face recognition.

SAR Image Compression Using Bounded Variation Component Analysis
Regis Marques, Daniel S. Ferreira, Fátima Medeiros, Janaina Cruz, and Maria Duarte
(pp 13 - 14). (paper)

This paper proposes a scheme to compress SAR (Synthetic Aperture Radar) images aiming for preserving fine details such as bright targets. This scheme consists in decomposing a SAR image into bounded variation (BV) and oscillating components. The adaptive compression scheme is a modified version of the well known SPIHT method named in this paper as MSPIHT. The experiments using real SAR images were compared with JPEG2000 and the conventional SPIHT. The assessment results indicated a superior performance of method in preserving fine details.

A Fourier Transform-based Method to Fusion IKONOS Data
Juliana Denipote (USP) and Maria Stela V. de Paiva (EESC-USP)
(pp 15 - 16). (paper) (poster)

The principal objective of image fusion in remote sensing is to obtain high-resolution multispectral images that can combine the spectral characteristic of the low-resolution multispectral bands with the spatial information of the high-resolution panchromatic band. A method based on Fourier transform is proposed in order to obtain good spatial and spectral resolutions using all bands of IKONOS satellite. Quantitative measurements were applied to compare the results of the proposed method with IHS, IHS enhanced by Fourier transform and wavelet-based methods.

A new Content-Based Image Retrieval Methodology Based on Image Registration Algorithms Processed in Grid Computing
Marcelo Oliveira (UFAL), Walfredo Cirne (UFCG), José Flávio Júnior (UFCG), and Paulo de Azevedo-Marques (USP)
(pp 17 - 18). (paper)

The Content-Based Image Retrieval (CBIR) has received great attention in the medical community because it is capable of retrieving similar images that have known pathologies. This work proposes a new methodology based on higher processing provided by the Grid Computing (GC) technology to achieve the CBIR using Registration algorithms. We evaluate the accuracy in retrieving images of the same plane, weighted and acquisition sequence related to a reference image. This works uses a heterogeneous image database and uses two anatomic regions (breast and head) as reference. The IR procedure uses Mean Square Metric (MSM) and Cross Correlation (CC). Both metrics showed a higher efficiency, MSM obtained 83% (breast) and 94% (head) precision average, the CC showed 81% (breast) and 98% (head). The higher computational cost related to the Image Registration were amortized by the GC, showing a potential low cost solution for secure data interchanging and integrating multiple hospitals and clinics.

Improving Content-based Image Retrieval by Combining Growing Hierarchical Self Organizing Map Classifiers for Color, Shape and Texture Features
Saulo de Tarso Rodrigues, Tiago Albuquerque, and Herman Gomes (UFCG)
(pp 19 - 20). (paper) (poster)

This paper presents the development of a content-based image retrieval system that combines a set of GHSOM (Growing Hierarchical Self Organizing Map)[4] classifiers taking as input color, shape and texture features. Experiments demonstrated combination scenarios that produced better accuracy in image classification.

Surface Equalization
Francisco B. Murilo (University of Costa Rica), Douglas N. Guevara (Universidad Nacional de Costa Rica), Francisco Torres-Rojas (Instituto Tecnológico de Costa Rica), and Alexander R. Gonzalez (University of Costa Rica)
(pp 21 - 22). (paper)

Subdivision is a method to create smooth surfaces through a refinable schema of polygonal or triangular meshes. From a mathematical point of view, this technique is an application of a second generation wavelet analysis. At the same time, this analysis is based on the lifting schema and does not use any frequency criterion to define scaling functions or wavelets. This simplifies computational cost. Yet, a frequency analysis applied to a polygonal mesh, can provide an intuitive method to modify surface characteristics by creating an analogy to one-dimensional sound equalization. These foundations can be useful not only as a surface generation tool, but they are also naturally associated to finite element techniques and can be applied to a wide variety of simulation problems.

Efficient Computation of Global Illumination for Image Synthesis
Anamaria Gomide, Danillo Pereira, and Jorge Stolfi (UNICAMP)
(pp 23 - 24). (paper) (poster)

We are exploring various functional approximation schemes (such as radial bases and Shepard’s interpolation) for precomputed radiance transfer (PRT), a finite element approach for real-time radiosity of complex scenes.

Using Metaprogrammed Functors to Implement Double-Dispatch for Collision Handling
Tiago Nobrega, Diego Carvalho, and Aldo von Wangenheim (UFSC)
(pp 25 - 26). (paper) (poster)

We provide a strategy to the management of collisions between multiple objects with different types employing multiple-dispatch and both Object-Oriented and Generic Programming concepts. The solution scales well to the number of object types, with a fixed, constant-time cost to arrive at the proper interference-detection routine. Additionally, it helps the application programmer by allowing them to implement only the functionality required by their own program, removing the need for common solutions such as Abstract Base Classes full of pure virtual methods.

A Geometric Bisection Algorithm for Automatic Page Layout
Alexandre Seki, Mauricio Klein, Ricardo Piccoli, Isabel Manssour, and João Oliveira (PUC-RS)
(pp 27 - 28). (paper) (poster)

Content distribution in a page designed for printing is a difficult task performed by human editors. There are applications and tools to help the editor in this task, but it is still a manual, iteractive process that is made more difficult when content is distributed in columns, as is the case of newspapers. This work describes an algorithm to automatically distribute content in a printable page. Our implementation tries to use all the page, dividing it among the items to be placed according to their area.

Enabling Interactive Brain Fiber Tracking with the GPU
Adiel Mittmann, Tiago Nobrega, Eros Comunello, Diego Carvalho, and Aldo von Wangenheim (UFSC)
(pp 29 - 30). (paper) (poster)

Brain fiber tracking allows a neurosurgeon to find fiber bundles in any region of the human brain, rendering it a useful aid to preoperative planning. However, this process requires a significant amount of computation, which limits the interactivity of current applications. We show how executing this process on graphics processing units can significantly improve the user’s experience by computing the fiber tracts in real time.

Physically Based Simulation Using Particle Systems
Yalmar Atencio (COPPE/UFRJ) and Claudio Esperança (UFRJ)
(pp 31 - 32). (paper)

This work consists of extending the rigid body simulation method presented by Harada. Firstly, the Newtonian formulation is replaced by impulse based physics. Secondly, deformable body simulation is achieved by using two different approaches: non-zero volume objects are handled by an adapted meshless shape matching technique, whereas objects such as cloths and ropes are simulated using Jakobsen’s approach.

3D Texture Painting of Point Models
José Ricardo M. Viana, Ricardo Marroquim, and Claudio Esperança (UFRJ)
(pp 33 - 34). (paper) (poster)

We propose a technique for painting 3D models based on a rendered image of the model and a 2D map of normal vectors. Texture generation and coordinate mapping is then produced on-the-fly, making it possible to paint in broad strokes or in small detail. The minimalistic requirements of this technique make it suitable for painting both regular meshes and point-based models.

Level of Detail for Point Model Rendering
Felipe Carvalho, Antonio Oliveira, and Ricardo Marroquim (UFRJ)
(pp 35 - 36). (paper)

Point-Based representation became a popular alternative to polygonal meshes for representing 3D geometric models. 3D photography and scanning systems acquire the geometry and appearance of real-world objects as point samples. In this work we present a method for efficiently creating a hierarchical multiresolution structure for point models. A variant of the octree is used to partition the space, while the merging of samples is driven by two error metrics.

Frature Pattern for Non-Photorealistic Image Rendering
Anna Regina Corbo and Luiz Henrique de Figueiredo (IMPA)
(pp 37 - 38). (paper)

We describe a method for non-photorealistic, “artistic” rendering of images that uses lines to depict fracture (or crack) patterns based on image features. The method starts by sampling the image with a set of points that is good for stippling. These points are then used as vertices of a minimum spanning tree which guides the crack propagation in image space.

Reviewers