Making the Most of Your Image: Colour and Statistics

There have been a number of interesting and exciting developments in computer vision research over the past decade. In this paper we briefly review some of those in which the contribution of British research groups has been prominent. Examples include model based vision, the application of invariance theory, active vision, and the application of statistical techniques to the modelling and interpretation of images of objects whose shape and appearance can vary from instance to instance. The main focus of the paper is on reviewing some recent developments in the application of statistical techniques, both at the pixel and object levels. The former is illustrated by examples from our own work on the modelling of colour images, whilst the latter is mainly concerned with the modelling of the shape and appearance of objects using principal components analysis as pioneered by the group at the University of Manchester in their Point Distribution and Active Shape Models. It is emphasised throughout how the use of such modelling techniques is dependent on the identification of problems for which there is a well defined ensemble of possible images leading to well defined distributions and manageable statistics. The importance of using up to date tools, programming methods and computing environments is illustrated in passing.