Marcelo Walter

Spatial sorting

Many graphics and also non-graphics applications need efficient techniques to find the nearest neighbors of a given query point. There are two approaches to address this problem: space-partitioning and data-partitioning. We present a data-partitioning error-controlled strategy for solving the nearest neighbor search (NNS) problem using spatial sorting as the basic building block. We improve on the neighborhood grid method by doing an extensive study on novel spatial sorting strategies for bidimensional NNS, providing significant performance and precision gains over previous works. Experiments demonstrate that, for many dense 2D point distributions, our solution is competitive with more complex and traditional techniques, such as k-d trees and index sorting. We also show comparable results for the 3D case. Our primary contribution is a dynamic, simple to implement, memory efficient, and highly parallelizable solution for low-dimensional approximate nearest neighbor search. Graphical abstractThis paper discusses in depth the problem of spatially sorting points in 2D and 3D, aiming to perform a fast, dynamic, and memory-efficient approximate nearest neighbor search.Display Omitted HighlightsWe analyze in depth the problem of spatially sorting points in two and three dimensions.We present novel and efficient spatial sorting algorithms.The performance and precision for nearest neighbor search (NNS) is evaluated.We show that spatial sorting is competitive against current NNS techniques.

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 pattern and a body shape. We present an approach that integrates a biologically-plausible pattern generation model, which can effectively deliver a variety of patterns characteristic of mammalian coats, and a body growth and animation system that uses experimental growth data to produce individual bodies and their associated patterns automatically. We use the example of the giraffe to illustrate how our approach takes us from a canonical embryo to a full adult giraffe in a continuous way, with results that are not only realistic looking, but also objectively validated. The flexibility of the approach is demonstrated by examples of big cat patterns, including an interpolation between patterns. The approach also allows a considerable amount of user control to fine-tune the results and to animate the resulting body with the pattern.

Towards local control for image-based texture synthesis

New advances in image based texture synthesis techniques allow the generation of arbitrarily sized textures based on a small sample. The generated textures are perceived as very similar to the given sample. One main drawback of these techniques, however, is that the synthesized result cannot be locally controlled, that is, we are able to synthesize a larger version of the sample but without much variation. We present in this paper a technique which improves on current fast texture synthesis techniques by allowing local control over the result. By local control we mean a final texture that is still perceived as a whole but presents variations in size of the basic elements. Our solution generates the final texture from a small collection of the same sample at different resolutions, adequately interpolated. We illustrate our results with some examples, including natural textures such as animal coat patterns, which exhibit local variations that can be adequately captured by our algorithm.

Modeling the structure of feathers

We present a parametric model for feather modeling in computer graphics. The model is based on Bezier curves and allows easy generation of many feather structures through manipulation of the parameters.

Growing and Animating Polygonal Models of Animals

While there exist many computer models of animal bodies, as polygonal meshes and parametric surfaces, these are difficult to modify to take growth into account, or to animate. Growth data available from the literature usually is expressed as very sparse measurements over the body at various ages of the animal. We present here basic techniques to transfer growth data to computer models (especially polygonal meshes), which allows animation of the growth as well as animation of the body in the traditional sense.

Patch-Based Texture Synthesis Using Wavelets

Patch-based texture synthesis builds a texture by joining together blocks of pixels — patches — of the original sample. Usually the best patches are selected among all possible using a L2 norm on the RGB or grayscale pixel values of boundary zones. The L2 metric provides the raw pixel-to-pixel difference, disregarding relevant image structures — such as edges — that are relevant in the human visual system and therefore on synthesis of new textures. We present a wavelet-based approach for selecting patches for patch-based texture synthesis. For each possible patch we compute the wavelet coefficients for the boundary region and pick the patch with the smallest error computed from the wavelet coefficients. We show that the use of wavelets as metric for selection of the best patches improves texture synthesis for samples which previous work fails, mainly textures with prominent aligned features.

3D mosaics with variable-sized tiles

Three dimensional mosaics, or surface mosaics, are a beautiful art form where a sculpture is made from putting together tiles on a given shape. Although 2D mosaics have attracted a lot of attention in graphics research, the same is not true for 3D mosaics. We present in this paper a technique to simulate 3D mosaics where the size of the individual pieces vary according to the local geometry. Previous work on this topic have addressed the problem of mosaics with tiles of the same size. The user specifies the number of tiles and our system covers a given 3D shape with square-shaped tiles with size adjusted according to the local curvature of the surface. The user can also controls the number of tiles, orientation on the surface, and relative size of tiles.

Selective hole-filling for depth-image based rendering

One of the biggest challenges in view interpolation is to fill the regions without projective information in the synthesized view. In this paper, we present a new approach that identifies and corrects different types of missing information. In the first stage, we propose a fast solution to tackle the problems of cracks and ghost, common artifacts in the view interpolation process. Then, we complete larger holes by exploring the disparity map as an additional cue to select the best patch in a patch-based inpainting procedure. Our experimental results indicate that we were able to outperform current state of the art hole filling techniques for view interpolation.

Virtual woodcuts from images

We present in this paper a technique for synthesizing virtual woodcuts based on real images. Woodcuts are an ancient form of art in which an image is printed from a block of carved wood. Our solution is fully automatic, but it also allows a great deal of user control if desired. Given an input image, our solution will synthesize an artistic woodcut from this image. Previous work on this topic has mainly relied on simulation of the actual physical interaction process between wood, ink, and paper, whereas we present a consistent solution based on four steps: image segmentation, computation of orientation fields, generation of strokes, and final rendering. Although some non-photorealistic rendering work could be possibly extended to approximate woodcut effects, ours is the first consistent approach targeting woodcuts specifically for images. We illustrate the potential of the proposed technique with examples of virtual woodcuts, obtained either automatically or user guided.

Synthesis and Transfer of Time-Variant Material Appearance on Images

The modeling of weathering effects in still images is a powerful editing tool for many graphics applications. Although most of these effects can be generated by manually creating textures through graphics editing applications, performing such artist work is time consuming. More sophisticated and automatic solutions to simulate a wide variety of weathering effects become essential. We present a method to generate and transfer weathering effects on images based on chroma distribution and luminance manipulation. Our solution has three main contributions over prior work. First, we propose an alternative solution to appearance manifolds, called appearance maps, to help synthesize variations in appearance in objects due to weathering effects. Although powerful, appearance manifolds are costly to build. The use of appearance maps allows real-time editing of images. Second, we extend the technique by using texture mapping ideas to allow the inclusion of weathering effects on images without them. Finally, we provide a mechanism to properly handle highlights, extending the range of possible images for manipulation. The results illustrate the flexibility of our solution and we also present results comparing real and virtual decay processes. Applications of this technique include virtual restorations of objects and images, as well simulation studies on future appearance of objects under environmental influence.