Publications

Performance and Quality Analysis of Convolution-Based Volume Illumination

P. Angelelli and S. Bruckner

Abstract

Convolution-based techniques for volume rendering are among the fastest in the on-the-fly volumetric illumination category. Such methods, however, are still considerably slower than conventional local illumination techniques. In this paper we describe how to adapt two commonly used strategies for reducing aliasing artifacts, namely pre-integration and supersampling, to such techniques. These strategies can help reduce the sampling rate of the lighting information (thus the number of convolutions), bringing considerable performance benefits. We present a comparative analysis of their effectiveness in offering performance improvements. We also analyze the (negligible) differences they introduce when comparing their output to the reference method. These strategies can be highly beneficial in setups where direct volume rendering of continuously streaming data is desired and continuous recomputation of full lighting information is too expensive, or where memory constraints make it preferable not to keep additional precomputed volumetric data in memory. In such situations these strategies make single pass, convolution-based volumetric illumination models viable for a broader range of applications, and this paper provides practical guidelines for using and tuning such strategies to specific use cases.

P. Angelelli and S. Bruckner, "Performance and Quality Analysis of Convolution-Based Volume Illumination," Journal of WSCG, vol. 23, iss. 2, p. 131–138, 2015.
[BibTeX]

Convolution-based techniques for volume rendering are among the fastest in the on-the-fly volumetric illumination category. Such methods, however, are still considerably slower than conventional local illumination techniques. In this paper we describe how to adapt two commonly used strategies for reducing aliasing artifacts, namely pre-integration and supersampling, to such techniques. These strategies can help reduce the sampling rate of the lighting information (thus the number of convolutions), bringing considerable performance benefits. We present a comparative analysis of their effectiveness in offering performance improvements. We also analyze the (negligible) differences they introduce when comparing their output to the reference method. These strategies can be highly beneficial in setups where direct volume rendering of continuously streaming data is desired and continuous recomputation of full lighting information is too expensive, or where memory constraints make it preferable not to keep additional precomputed volumetric data in memory. In such situations these strategies make single pass, convolution-based volumetric illumination models viable for a broader range of applications, and this paper provides practical guidelines for using and tuning such strategies to specific use cases.
@ARTICLE {Angelelli-2015-PQA,
author = "Paolo Angelelli and Stefan Bruckner",
title = "Performance and Quality Analysis of Convolution-Based Volume Illumination",
journal = "Journal of WSCG",
year = "2015",
volume = "23",
number = "2",
pages = "131--138",
month = "jun",
abstract = "Convolution-based techniques for volume rendering are among the fastest  in the on-the-fly volumetric illumination category. Such methods,  however, are still considerably slower than conventional local illumination  techniques. In this paper we describe how to adapt two commonly used  strategies for reducing aliasing artifacts, namely pre-integration  and supersampling, to such techniques. These strategies can help  reduce the sampling rate of the lighting information (thus the number  of convolutions), bringing considerable performance benefits. We  present a comparative analysis of their effectiveness in offering  performance improvements. We also analyze the (negligible) differences  they introduce when comparing their output to the reference method.  These strategies can be highly beneficial in setups where direct  volume rendering of continuously streaming data is desired and continuous  recomputation of full lighting information is too expensive, or where  memory constraints make it preferable not to keep additional precomputed  volumetric data in memory. In such situations these strategies make  single pass, convolution-based volumetric illumination models viable  for a broader range of applications, and this paper provides practical  guidelines for using and tuning such strategies to specific use cases.",
pdf = "pdfs/Angelelli-2015-PQA.pdf",
images = "images/Angelelli-2015-PQA.jpg",
thumbnails = "images/Angelelli-2015-PQA.png",
keywords = "volume rendering, global illumination, scientific visualization, medical visualization"
}
projectidprojectid

Media

Downloads

Full paper [PDF]