This dissertation offers an explanation of why and how maximum a posteriori (MAP) image restoration works. A new numerical iteration scheme is derived on the basis of this explanation with a convergence criterion related to the noise process. This numerical scheme is very similar to the modified Picard's method given in standard numerical analysis texts. The effects of the parameters of the MAP restoration method are predicted by the explanation and confirmed by experiment.nThe MAP restoration method is further improved by using local processing, i.e., processing small sections of the image sequentially and piecing them together to form the restored image. It was found, as predicted, that smaller section sizes result in better restorations. Computing costs increase as section size decreases.
展开▼
