This project deals with three related problems in the general area of image processing. The first is a dramatic increase in the efficiency of the Gibbs Sampler and Metropolis Dynamic algorithms from restricting the algorithms to a locally bounded image space. The second develops a special estimation method for astronomical image data of a faint disk, where the data has significant blur (e.g., Hubble telescope data). Here the use of a suitable lower dimensional linear function improves efficiency. The third addresses data compression using a Markov random field model and subdividing the original large image and forming optimal representative subimages. Statistical image analysis uses statistical principles to "guess" the real image when only poor quality image data is available. The quality of a particular algorithm is depends upon how good the chosen "guess" is and how quickly it is achieved. This project focusses on increasing the speed of the algorithms by trying to get the most pertinent information rather than working on the whole image at once.