Abstract:
This thesis considers the problem of identification and restoration of images degraded by additive Gaussian white noise. It is assumed that the power of the noise and the statistical properties of the original image are not known a priori. A new approach which reduces the two dimensional problem to a one dimensional problem by using the unitary discrete Fourier transform is introduced. Then, by applying the expectation-maximization (EM) algorithm, the image is restored and the parameters of various types of AR models are identified under noisy conditions. Two different methods are used for restoration, namely, maximum likelihood restoration and Kalman filtering. The simulation results of the presented approach are also included.
