TABLE OF CONTENTS Since any real image recording system must use a finite aperture and is subject to measurement noise, we are naturally interested in methods of correcting imperfections in the recorded image. The goal of image restoration is to estimate the original appearance of an image that has been degraded in some way. Typical degradations include optical blurring, geometric warping, sensor transformations, quantization and electronic noise. Even halftoning in the reproduction process can be considered a degradation that may be corrected by some later operation. It is important to distinguish between restoration and enhancement, which also tries to improve the appearance of an image. The restoration methods discussed here are applied to both spatial degradations (blurring and noise) and spectral degradations (color distortions and noise).
Restoration is a mathematically well defined process that estimates the original signal from recorded data. To restore an image much quantitative information is required, in particular:
Image formation model with known parameters. The models and their parameters are discussed in Chapter 13. Of course, the accuracy with which the parameters are known affects the quality of the restoration.
Restoration criterion. This is usually an optimization criterion of some type. The minimum mean square error is the one most frequently used. This is done for ease of analysis and mathematical tractability. However, with the advent of faster computers, more visually meaningful cost functions are being used, such as 
for color images. In addition, constrained problems can be formulated that account for a priori knowledge...
