Abstract
The distribution of colors in an image often provides a useful cue for image indexing and object recognition. However, two problems are reported in the literature: firstly, color distributions are dependant on the illumination color, and secondly, that color distributions represented as histograms are large in size - up to 4000 bins - thus limiting the scale of the database that might reasonably be indexed. Both of these problems have been separately addressed in the literature. However, the derived solutions are not compatible with one another. In this paper, we look at both problems together and at the same time. We develop a parsimonious representation, which consists of distinct illuminant dependent and independent parts.Our representation is based on a log-opponent chromaticity representation. By using chromaticities, we avoid the problem of brightness indeterminacy (the power and position of the light source). Opponency gives a perceptually relevant and efficient coding. Finally, the use of logarithms renders illuminant change simple to model: as the illumination changes, so the distribution of log-opponent chromaticities undergoes a simple translation. We code log-opponent chromaticity distributions by the distribution mean and the lowest k statistical moments (where the moment functions are discovered using Principle Component Analysis). We show that only the mean in this expansion depends on illumination. Experiments show two important results - indexing using both mean and as few as 8 moments delivers near perfect indexing for an illuminant color corrected database (removing the mean leads to slight degradation of performance), while indexing without the mean delivers near perfect indexing for Funt et al's illuminant dependent images (poor indexing results if the mean is used).