Nonrecursive digital image restoration
M. J. McDonnell · UC Research Repository (University of Canterbury) · 1975
The problem of digitally restoring blurred images is considered. The subject of image restoration is reviewed in detail and a comprehensive notation for image restoration is developed. Degraded images are divided into classes G and S - those which are and are not respectively truncated by their recording frames. It is shown that conventional restoration techniques only work well for images of class S. Restoration techniques for dealing with class G images are presented, as well as improved techniques for dealing with class S images. These new techniques are shown to be both efficient and practical through the use of computer simulations, and through the restoration of optically degraded images. The problem of designing an optimum nonrecursive digital filter array of a given number of elements is solved in both one and two dimensions. It is demonstrated that this new design method is superior to previous techniques. A sampling theory appropriate for deconvolution problems is presented. New sampling functions are introduced, in both one and two dimensions, which overcome the "picket fence" effect associated with the sine sampling function. The concept of a "line-segment-limited" function is defined. It is shown that a pseudo noise level is introduced by the approximation that the point spread function is line-segment-limited. It is this pseudo noise level, and not aliasing errors, that restricts the choice of a sampling rate less than the Nyquist rate. New criteria are presented which allow the choice of a sampling rate considerably less than the Nyquist rate.