doi:10.1016/S0097-8493(97)00084-8
Copyright © 1997 Elsevier Science Ltd. All rights reserved
Scene Simplification
Smooth transitions in texture-based simplification
D. G. Aliagaa, * and A. A. Lastraa
a Department of Computer Science, University of North Carolina, Chapel Hill, NC 27599-3175, USA
e-mail: {aliaga | lastra}@cs.unc.edu
Available online 30 November 1998.
References and further reading may be available for this article. To view references and further reading you must
purchase this article.
Abstract
We are investigating techniques for providing smooth transitions when simplifying large, static geometric models with texture-based representations (or impostors). Traditionally, textures have been used to replace complex geometry, for example the books on a shelf or the complex foliage on a tree. Using textures in a more general manner is a relatively new area of research. The key idea is that 2D image textures can be used to temporarily represent 3D geometry. Rendering speed is increased if a replacement texture can be used for several frames, since textures can be rendered quickly and independently of model complexity. Because a texture is only correct from a single viewpoint, visual discontinuities, such as misalignment with adjacent geometry, begin to appear as the texture is used from other viewpoints. Previous approaches have controlled these errors by re-rendering the textures often or providing a large set of precomputed textures. We have improved upon these methods by developing algorithms for: (a) providing continuous imagery across borders between geometry and sampled textures at all times; (b) providing smooth dynamic transitions between geometry and texture.
Author Keywords: geometry, textures, warping, morphing, visual complexity, space partitioning, simplification, visibility culling, interactive
Index Terms: Computational geometry; Image quality; Image reconstruction; Computer simulation; Three dimensional computer graphics; Anti-aliasing; Error analysis; Computational complexity; Image warping; Image morphing; Visual complexity; Space partitioning; Visibility culling
Fig. 1. Church model. Geometry around textures (outlined in red) is warped.
Fig. 3. No geometry warping (texture outlined in red). (a,b) Eye moves away from texture-viewpoint. (c) For clarity, geometry is rendered in wireframe.
Fig. 5. Texture-to-Geometry Transition. (a) Texture (outlined in red) is about to be returned to geometry. (b) Midpoint of morphing sequence. (c) Geometry has been fully restored.
Fig. 4. Geometry warping. (a,b) Eye moves away from texture-viewpoint. Notice the geometric continuity at the boundary. (c) For clarity, geometry is rendered in wireframe.
Fig. 7. Sequence of transformations for geometry warping. We first project along di onto the plane of the quadrilateral using the interpolated view frustum or the texture view frustum (in the latter case, pi=pa, Pi=Pa and Mi=Ma).
Fig. 8. Common viewpoint, adjacent textures. (a) Textures at equal view depth, thus will maintain geometric continuity. (b) Varying view depth but same at the seams, geometric continuity maintained. (c) Discontinuous view depth, geometric continuity not maintained. This case can easily be avoided.
Fig. 9. Smooth transitions. (a) Removing a texture at the edge of a string of adjacent textures, interpolation required. (b) Removing a texture in the middle of a string of adjacent textures, no interpolation.
Fig. 10. Multiple textures. Different viewpoints and distances.
Fig. 6.
Model partitioning for geometry warping. Each box corresponds to a space-partitioning box. The boxes are classified: near, texture and surrounding. Intersected boxes can be optionally partitioned.
Fig. 11. Warping sequence. A geometry-to-texture transition goes from (a) to (c). At the end of the transition, the texture is introduced. A texture-to-geometry transition goes from (c) to (a). From the texture-viewpoint, the objects look the same at all times.
Fig. 12. (a) Portal Culling. (b) Portal Texture Culling.
Fig. 2. Brooks House model. Portals have been replaced with textures.
Table 1. Performance summary
