Color has been widely used in content-based image retrieval applications. In such applications the color properties of an image are usually characterized by the probability distribution of the colors in the image. A distance measure is then used to measure the similarity between images based on the descriptions of their color distributions. In this thesis we develop statistical methods which focus on the representation of such distributions and distance measures between distributions.

The distance measures are described in a differential geometry based frame­ work. This allows the incorporation of geometrical features of the underlying color space into the distance measure between the probability distributions. The general framework is illustrated with two examples: Normal distributions and linear representations of distributions. The linear representation of color distributions is used to derive new compact descriptors for color based image retrieval. These descriptors are based on the combination of two ideas: Integration of color information into the principal component computation and using local differences of histograms instead of all histograms for the estimation of the principal components. In our experiments we used several image databases containing about 200000 images. The experiments show that the method developed in this thesis is very fast and that the retrieval performance achieved compares favorably with existing methods.

We also describe illumination invariant descriptors that can be used in, image database searches which retrieves images of objects independent of the illumination conditions under which these images were taken. We develop and investigate a moment based method for color image normalization and compare it to traditional color constancy methods.

In the last chapter of the thesis we describe an industrial application of some of the ideas described in the first part of the thesis. In this application we use color correction methods to optimize the layout of a newspaper page.