From the Rendering Equation to Stratified Light Transport Inversion

  • Authors:

  • Tian-Tsong Ng;Ramanpreet Singh Pahwa;Jiamin Bai;Kar-Han Tan;Ravi Ramamoorthi

  • Affiliations:

  • Institute for Infocomm Research, Singapore, Singapore 128632;Department of Electrical and Computer Engineering, University of Illinois at Urbana-Champaign, Urbana, USA 61801-2918;EECS Department, University of California at Berkeley, Berkeley, USA 94720-1776;HP Labs, Palo Alto, USA 94304;EECS Department, University of California at Berkeley, Berkeley, USA 94720-1776

  • Venue:
  • International Journal of Computer Vision
  • Year:
  • 2012

Quantified Score

Hi-index 0.00

Visualization

Abstract

Recent advances in fast light transport acquisition have motivated new applications for forward and inverse light transport. While forward light transport enables image relighting, inverse light transport provides new possibilities for analyzing and cancelling interreflections, to enable applications like projector radiometric compensation and light bounce separation. With known scene geometry and diffuse reflectance, inverse light transport can be easily derived in closed form. However, with unknown scene geometry and reflectance properties, we must acquire and invert the scene's light transport matrix to undo the effects of global illumination. For many photometric setups such as that of a projector-camera system, the light transport matrix often has a size of 105脳105 or larger. Direct matrix inversion is accurate but impractical computationally at these resolutions.In this work, we explore a theoretical analysis of inverse light transport, relating it to its forward counterpart, expressed in the form of the rendering equation. It is well known that forward light transport has a Neumann series that corresponds to adding bounces of light. In this paper, we show the existence of a similar inverse series, that zeroes out the corresponding physical bounces of light. We refer to this series solution as stratified light transport inversion, since truncating to a certain number of terms corresponds to cancelling the corresponding interreflection bounces. The framework of stratified inversion is general and may provide insight for other problems in light transport and beyond, that involve large-size matrix inversion. It is also efficient, requiring only sparse matrix-matrix multiplications. Our practical application is to radiometric compensation, where we seek to project patterns onto real-world surfaces, undoing the effects of global illumination. We use stratified light transport inversion to efficiently invert the acquired light transport matrix for a static scene, after which interreflection cancellation is a simple matrix-vector multiplication to compensate the input image for projection.