Purchase this article with an account.
Katja Doerschner, Huseyin Boyaci, Laurence T. Maloney; Representing spatially and chromatically varying illumination using spherical harmonics in human vision. Journal of Vision 2005;5(8):785. doi: 10.1167/5.8.785.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
When light sources are distant, they can be represented as a spectral power distribution on a Debevec sphere (DS; Debevec & Malik, SIGGRAPH, 1997). In natural scenes, this distribution can be complex, and the light absorbed by a matte surface can vary with surface orientation. We previously reported that observers discount orientation changes in 3D scenes lit by a diffuse blue and a punctate yellow source (Boyaci et al, JOV, 4, 664–679). Observers effectively estimated at least some aspects of the DS.
In a spherical harmonics expansion of the DS, only the first nine low-pass components contribute appreciable light to matte surfaces (Basri & Jacobs, IEEE/PAMI, 2003). Here we examine whether the visual system makes use of this physical constraint to simplify representation of illumination. We asked observers to carry out an achromatic setting task for matte patches varying in orientation with full-pass and matched low-pass DSs. If performance were the same in full-pass and matched low-pass scenes, we could conclude that the visual system uses only the low-pass.
Methods: Stimuli were binocularly-viewed rendered 3D scenes. We chose four DSs, two full-pass consisting of two yellow punctate light sources and a blue diffuse source, and two corresponding low-pass approximations with nine components. Test patch orientation was varied and observers adjusted the color of the test until it was perceived to be achromatic; 7 naive observers repeated 9 orientations and 4 DS conditions 20 times.
Results: Under the low pass approximations observers did not compensate as well for changes in orientation. We conclude that the visual system makes use of high-pass information in estimating the DS. We propose a two-stage model: the visual system estimates the illumination using its full spatial spectrum, and then at a second stage retains only the low-pass representation for the illumination in estimating the surface color of matte surfaces.
This PDF is available to Subscribers Only