Abstract
This paper presents image modeling and restorationby higher-order statistics based 2-D inverse filters. Agiven original image +(m, n) is processed by an optimuminverse filter v(m, n) which is designed by maximizingcumulant based criteria Jr,m = ICmlr/ICrlmwhere r is even, m > T 2 2 and C,,, (Cr) denotes mthorder(rth-order cumulant of the output e(m,n) ofbe modeled as the output of a linear shift-invariant(LSI) system h(m, n) driven by e(m, n) where h(m, n)is a stable inverse filter of v(m,n). When a blurredimage y(m,n) = t(m,n) * g(m,n,) rather than theoriginal image z(m, n) is given, t(m, n) can be restoredby first estimating e(m,n) using the previousinverse filter criteria and then obtain t(m,n) =e(m, n) * h(m, n). Some experimental results are providedto support the proposed image modeling andrestoration method.