A new Trust Region Fisher Scoring optimization for image and blur identification

In this paper, we present a Fisher Scoring (FS) based method that ensures the convergence in every iteration step to identify the image parameters (blur coefficients and image model parameters). We analyze the Trust Region (TR) version of FS and we describe a new optimization criterion. Thanks to the method, convergence to a stationary point is guaranteed if the step is in the “trust region”. This new algorithm has strong convergence properties and is robust and efficient in practice. Regularized solution is a side benefit of the proposed method.