Workshop on Image-Based Graphics

March 7-8, 2006

Organizers: Paulo Cezar Carvalho and Luiz Velho
Sponsored by Instituto do Milenio - AGIMB

Tuesday, March 7 (1:00 - 5:00)

13:00 - Opening
Luiz Velho and Paulo Carvalho, IMPA
13:30 - Modeling and Real-Time Rendering of Surface Details Using Relief Mapping
Manuel Oliveira, UFRGS
15:00 - Correspondences for Vision and Graphics
Siome Klein Goldenstein, UNICAMP

Wednesday, March 8 (10:30 - 5:00)

10:30 - Real time 3D video using (b,s)-BCSL code
Marcelo Bernardes, UFJF
13:30 - (Re-)Visiting Two Vison Problems
Michael Goesele, University of Washington
15:00 - Panel: Perspectives on Image-Based Graphics
All speakers

* All sessions will be held at IMPA, Auditorio 1

Modeling and Real-Time Rendering of Surface Details Using Relief Mapping
Manuel Oliveira, UFRGS

I will present three related image-based techniques for mapping surface details to polygonal models. First, I will describe a technique for mapping relief textures onto arbitrary polygonal models in real time. In this approach, the mapping of the relief data is done in tangent space. As a result, it can be applied to polygonal representations of curved surfaces producing correct self-occlusions, interpenetrations, shadows and per-pixel lighting effects. The approach can be used to consistently add surface details to geometric models undergoing deformations, such as in the case of animated characters commonly found in games. The technique uses an inverse formulation (i.e., pixel driven) based on an efficient ray-height-field intersection algorithm implemented on the GPU. It supports extreme close-up views of the surfaces, mip mapping and anisotropic texture filtering. Also, contrary to high-dimensional representations of surface details, the low memory requirements of the proposed technique do not restrict its use to tiled textures.

Next, I will show how to extend this first technique to render correct silhouettes. For this, each vertex of the polygonal model is enhanced with two coefficients representing a quadric surface that locally approximates the object's geometry at the vertex. Such coefficients are computed during a pre-processing stage using least-squares fitting and are interpolated during rasterization. Thus, each fragment contributes a quadric surface for a piecewise-quadric object-representation that is used to produce correct renderings of geometrically-detailed surfaces and silhouettes.

In the end, I will show how to extend the previous techniques to support the mapping of non-height-field surface details. It generalizes the notion of relief mapping to support multiple layers. This technique can also be used to render realistic impostors of 3D objects that can be viewed from close proximity and from a wide angular range. Contrary to traditional impostors, these new one-polygon representations can be observed from both sides, producing correct parallax and views that are consistent with the observation of the 3D geometry they represent.

This work was done in collaboration with Fabio Policarpo and Joao Comba

Correspondences for Vision and Graphics
Siome Klein Goldenstein, UNICAMP

Nesta palestra, abordamos a importancia do problema de estabelecer correspondencias entre imagens tanto para visao como para computacao grafica. Classificamos os tipos de situacoes para o calculo das correspondencias, e analisamos tres tecnicas ja estabelecidas para sua determinacao. Distinguimos os erros de precisao e erros grosseiros e como os aplicativos podem contorna-los. Finalmente, descrevemos e validamos o UKLT, uma nova tecnica facil de implementar e utilizar que desenvolvemos recentemente. Alem de calcular correspondencias, o UKL prove melhor informacao a respeito da precisao e elimina de forma mais eficiente erros grosseiros (outliers).

Real time 3D video using (b,s)-BCSL code
Marcelo Bernardes, UFJF

To be announced.

(Re-)Visiting Two Vison Problems
Michael Goesele, University of Washington

I will present results from two recent projects:

First, I will describe a simple and robust method for surface mesostructure acquisition. The method builds on the observation that specular reflection is a reliable visual cue for surface mesostructure perception. In contrast to most photometric stereo methods, which take specularities as outliers and discard them, the proposed progressive acquisition system captures a dense specularity field as the only information for mesostructure reconstruction. The method can efficiently recover surfaces with fine-scale geometric details from complex real-world objects with a wide variety of reflection properties, including translucent, low albedo, and highly specular objects.

In the second part of the talk, I will present an extremely simple yet robust multi-view stereo algorithm and analyze its properties. The algorithm first computes individual depth maps using a window-based voting approach that returns only good matches. The depth maps are then merged into a single mesh using a straightforward volumetric approach. I will present results for several datasets, showing accuracy comparable to the best of the current state of the art and rivaling more complex algorithms.