Abstract
In previous work (Logvinenko & Maloney, P&P, 2006), we asked observers to rate the dissimilarity of pairs of achromatic surfaces in simple scenes illuminated by neutral lights differing in intensity. We found that log surface albedo difference and log lighting intensity difference contributed additively to rated dissimilarity but that the surface contribution was greater, a form of partial lightness constancy. Here we (a) extend these results to dissimilarity of coloured surfaces under neutral illuminants differing in intensity and (b) compare two methods for measuring perceived dissimilarity, quadruple comparisons and dissimilarity ranking. In quadruple comparisons observers judge which of two illuminant-surface pairs exhibits the greater dissimilarity (forced choice). In dissimilarity ranking observers assign a numerical ranking to each illuminant-surface pairs.
Five normal trichromatic observers took part in the experiment. A stimulus display consisted of 3 identical sets of 7 Munsell papers (5R4/14, 5YR7/12, 5Y8/12, 5G6/10, 10BG5/8, 5PB5/12 and 10P5/12) illuminated independently by 3 light sources the intensities of which were 6, 110 and 2100 lux. The experiment was divided into two parts. In the first part, dissimilarities between the seven Munsell papers for each of the three illuminations were evaluated by using a method of quadruple comparisons. Each pair of pairs was evaluated five times for each observer. Then, in the second part, dissimilarities between Munsell papers illuminated by dimmer lights (6 and 110 lux, respectively) were evaluated by numerical ranking using the dissimilarities under the brightest light as anchors. We compared non-metric Multi-Dimensional Scaling solutions for the two methods. Procrustes statistics (a standard method for comparing spatial configurations) were close to zero. This justifies the use of the ranking method in measuring colour dissimilarity. We conclude that on average the dissimilarities between Munsell papers decreased with decreasing light intensity.
EPSRC research grant EP/C010353/1 (ADL) and NIH EY08266 (LTM)