Computational aspects of regularized image reconstruction

Image restoration procedures are commonly unstable in the presence of noise, and some technique for restoring stability becomes essential. The methods of regularization theory are particularly appropriate for this purpose. A specific type of regularized solution is introduced in the general context of image reconstruction. A super-resolution problem is then considered from the point of view of the computational tasks involved, with particular reference to the estimation of certain key parameters and to implementations which increase the efficiency of the calculations. Parameter estimation is performed by weighted cross-validation. The improvement in efficiency is achieved through the exploitation of symmetries or cyclic properties inherent in the reconstruction operator. The concept of displacement rank is introduced and estimates made of the computational burden associated with various classes of regularized reconstruction matrices.