Optimal image deconvolution by range and noise moment constraints
Angélica M. Díaz · 2005
Image deconvolution, also known as image restoration, is concerned with the estimation of an uncorrupted image from a noisy, degraded one. The degradation of this image may be caused by defects of optical lenses, nonlinearity of the electro-optical sensor, relative motion between an object and camera, wrong focus, etc. By assuming a degradation model, one can formulate and develop a restoration algorithm. In this thesis, the developed algorithms are iterative deconvolution methods based on noise moment and pixel range constraints. The moments were used to ensure that noise associated with the deconvolution solution satisfies predetermined statistics. The pixel range constraints were also used to ensure the solution is within predetermined pixel value bounds. This addresses the critical issue of noise amplification at those frequencies where the point-spread function (the blurring function) contains frequency nulls. The solution’s dependence on the number of moments is examined and the performance of the deconvolution approach is compared with existing and well established deconvolution methods such as Wiener filtering and inverse filtering.