ABSTRACT

The most obvious feature of digital images, considered as numerical data, is their very large size. Images very seldom consist of as few as 32 >< 32 pixels (picture elements): sizes of between 256 x 256 and 1024 x 1024 are much more common. And if a time dimension is added, through consideration of 'movies' or image sequences, the volume of data available becomes very large indeed. This size factor immediately favours the use of MCMC methods in a statistical approach to image analysis since although other more conventional numerical methods sometimes provide practical routes to the calculation of point estimates cf a true image, e.g. the maximum a posteriori (MAP) estimate., MCMC is usually the only approach for assessing the variability of such estimates.