Image enhancement via specification of moments of the intensity distribution

This dissertation considers the problem of image enhancement via specification of statistical moments of the intensity distribution of the image. A method is presented which first generates a distribution having the specified moments, then uses that distribution to process the image using intensity mapping techniques. Relationships between various moments and visual aspects of an image are established in order to provide a link between enhancement goals and the technique. This also aids user understanding of the enhancement process. For example, contrast is shown to be not only a function of variance, but also to be related to skew, especially for bimodal distributions. Higher moments are found to have less effect as the order of the moment increases. Specification of only the first few moments is shown to accomplish useful enhancement. Heuristic guide-lines are developed for selecting the specified moment values, given the moments of the original image. An algorithm is also presented which guides the selection of moment values in order to obtain a consistent set of moments with respect to probability density functions as defined here.