A new iterative algorithm for image restoration based on maximum likelihood principle

Considers the development and implementation of a new gradient-based algorithm for image restoration. The algorithm assumes that the original intensity signal s(x) has been affected by a known linear, but not necessarily space-invariant, point spread function in an additive white Gaussian noise environment. It is assumed that the covariance function of s(x) is known a priori. Based on these assumptions, the algorithm tends toward the maximum likelihood estimate of s(x) using the steepest ascent routine. The developed algorithm is reduced to the least squares error restoration scheme reported by E.S. Angel and A.K. Jain (1978) in the absence of noise when the covariance function of s(x) is an impulse function. Simulation experiments are presented