Lightness recovery for pictorial surfaces

Anna Paviotti, David Alexander Forsyth, Guido M. Cortelazzo

Research output: Contribution to journalArticle

Abstract

An important technique in cultural heritage preservation is multispectral acquisition, where one recovers a detailed spectral record of a painting using carefully calibrated lighting. This is difficult to do with frescoes, because it is hard to recover the spatial variation in light intensity that results from factors like the imaging setup and the curvature of the fresco. We introduce a new formulation of the lightness problem applied to images of pictorial artworks. The problem is different from the conventional lightness problem, because artists often paint the effects of light, so the albedo field contains a component that mimics an illumination field. Our method distinguishes between physical illumination and painted shading through spatial frequency effects and dynamic range considerations. We evaluate our method using multispectral images of paintings, where the physical illumination field is known. Our method produces estimates of the illumination intensity field that compare very well with the known ground truth, and outperforms other state-of-the art lightness recovery algorithms. For frescoes, ground truth is not available, but we show that our method produces consistent results, in the sense that the illumination functions estimated on the image and on (some of) its subimages are very similar on the overlap. We show our method produces qualitatively good color corrections for images of frescoes found on the web.

Original languageEnglish (US)
Pages (from-to)54-77
Number of pages24
JournalInternational Journal of Computer Vision
Volume94
Issue number1
DOIs
StatePublished - Aug 1 2011

Keywords

  • Color constancy
  • Color correction
  • Cultural heritage
  • Lightness problem
  • Multispectral imaging

ASJC Scopus subject areas

  • Software
  • Computer Vision and Pattern Recognition
  • Artificial Intelligence

Fingerprint Dive into the research topics of 'Lightness recovery for pictorial surfaces'. Together they form a unique fingerprint.

  • Cite this