Purchase this article with an account.
Sarah Allred, Vanessa Troiani, Lynn Lohnas, Li Jiang, Ana Radonjic, Alan Gilchrist, David Brainard; An ideal observer model predicts lightness matches. Journal of Vision 2009;9(8):345. doi: 10.1167/9.8.345.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Background: We seek general principles that allow prediction of perceived lightness for a large class of images using measurements of a small subset of images. Here we consider the class of grayscale checkerboard images.
Psychophysics methods: Observers viewed 25-square checkerboard images presented on a high-dynamic range computer display. Observers matched the lightness of the center square to grayscale surfaces presented in a separately illuminated booth. The remaining 24 squares defined the viewing context for the center square. For each context, we measured the full mapping between center square luminance and matched reflectance.
Model methods: We formulated a Bayesian algorithm that estimates surface reflectance and illuminant intensity. Algorithm estimates were driven by prior distributions over surface reflectance and illuminant intensity. Priors allowed both surface and illuminant properties to vary with spatial location, but contained a bias that favored more slowly varying illumination. The algorithm was converted to a model via the linking hypothesis that two checkerboard squares have the same lightness when the algorithm estimates their surface reflectance to be the same.
Results: A number of distinct factors of the checkerboard context were varied. These include highest square luminance, lowest square luminance, distribution of square luminances, and the spatial arrangement of the squares within the contextual checkerboard. All of these factors affected the mapping between central square luminance and matched reflectance. The psychophysical data were used to determine a small number of parameters that characterized the Bayesian algorithm's priors. The resulting model provided a good account of performance.
Conclusions: We present a quantitative model for lightness perception in arbitrary checkerboard scenes. Although the current model does not account for lightness phenomena that arise with more complex geometric arrangements, the surface-illuminant estimation principles that drive it are general, and thus have the potential to be elaborated to include such phenomena.
This PDF is available to Subscribers Only