September 2005
Volume 5, Issue 8
Free
Vision Sciences Society Annual Meeting Abstract  |   September 2005
Image statistics and reflectance estimation
Author Affiliations
  • Lavanya Sharan
    M.I.T. Dept. of Electrical Engineering and Computer Science
  • Yuanzhen Li
    M.I.T. Dept. of Brain and Cognitive Sciences
  • Edward H. Adelson
    M. I. T. Dept. of Brain and Cognitive Sciences
Journal of Vision September 2005, Vol.5, 375. doi:10.1167/5.8.375
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to Subscribers Only
      Sign In or Create an Account ×
    • Get Citation

      Lavanya Sharan, Yuanzhen Li, Edward H. Adelson; Image statistics and reflectance estimation. Journal of Vision 2005;5(8):375. doi: 10.1167/5.8.375.

      Download citation file:


      © 2016 Association for Research in Vision and Ophthalmology.

      ×
  • Supplements
Abstract

Humans are surprisingly good at judging the reflectances of complex surfaces even when they are viewed in isolation, contrary to the Gelb effect. We have previously argued that certain image statistics are useful in this task. We have now collected high dynamic range digital images, using multiple exposures, for a large number of materials, in multiple lighting conditions. After normalizing all images to have the same mean luminance, we used ROC analysis to evaluate the utility of various individual image statistics in classifying surfaces as black or white. We tested the statistical moments of the luminance histograms, i.e., variance, skew, kurtosis, and the 5th moment, and found them all to be useful. Extreme percentile statistics (e.g., 2nd and 98th percentile) were also useful. Since the human visual system is unlikely to have direct access to pixel luminances, we also considered the information available after filtering with simple filters such as a center-surround or oriented filter. The same statistics were useful when applied to the outputs of these filters. Single statistics can achieve classification rates of 70–80%. When human observers are asked to classify the images, they achieve similar rates, and tend to make errors on the same images that the machine does. This supports the notion that the image statistics capture information that is being used by humans.

Sharan, L. Li, Y. Adelson, E. H. (2005). Image statistics and reflectance estimation [Abstract]. Journal of Vision, 5(8):375, 375a, http://journalofvision.org/5/8/375/, doi:10.1167/5.8.375. [CrossRef]
Footnotes
 Supported by NTT Labs and MURI/ONR grant N00014-01-0625
×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×