Abstract
Temporal, or animation, antialiasing of algorithmic textures pose a particular challenge to the rendering of computer images because the high-frequency nature of some textures may often produces dramatically different results with only a small change in the sample point. The aliasing is particularly evident as a texture slowly moves across the computer screen during an animation, when it demonstrates the dreaded “shimmer”. This paper describes a simple and general method to limit a texture’s frequency as a function of screen pixel size.
Keywords: algorithmic textures, temporal antialiasing
1 Introduction
3D computer imagery is often colored by executing an algorithm to create resolution-independent simulations of wood, stone, scales, water, and a large variety of other natural materials [Perlin 85][Peac85]. These colors are mapped onto the model geometry per the 3D coordinate of the surface point. Screen coordinates, called pixels, are normally converted to a model space before being used by the texture algorithm.
In the generation of all digitally created computer images, antialiasing is the term used to describe the softening of visually objectionable jaggies that result when taking discrete samples from a mathematically continuous sample space, called spatial aliasing. In particular, texture generation algorithms are notoriously jaggy. The predominate method of spatial antialiasing is oversampling, which averages multiple, slightly offset texture samples with the original sample [Dippe & Wold 85][Lee et. al. 85][Kajiya 86][Mitchell 87]. Oversampling is either random, called stochastic sampling, or plotted from a fixed grid, but both techniques usually dynamically determine whether more samples are needed before executing additional time-consuming algorithmic texture computations. For still images that do not move, this solution works exceptionally well, but it fails during animation, which is called temporal aliasing. The underlying problem is the resolution-independent nature of algorithmic texture generation that unfortunately produces pathological situations where visually high-frequency attributes of the texture appear and disappear from frame-to-frame. They shimmer, even though these high-frequency attributes are spatially antialiased on any particular frame. As an example, assume the area to be covered is actually very large in model space but is reduced to one pixel of screen space due to perspective projection. Twenty samples randomly averaged from this large model space area can often return a completely different color than twenty other random samples from the same area. As the sample area moves on the screen, each frame may generate twenty new random samples, so the texture shimmers.
2 Bitmap Textures
Published in "Art+Math=X" Conference at the University of Colorado - Boulder, June 2-5, 2005
General Algorithmic Texture Antialiasing 2005
-
- Posts: 18752
- Joined: Wed Jan 20, 2010 2:02 pm
General Algorithmic Texture Antialiasing 2005
You do not have the required permissions to view the files attached to this post.
Shamedia, Shamdemic, Shamucation, Shamlection, Shamconomy & Shamate Change