Model identification and parameters estimation for image deblurring

Abstract

Blurred images usually come from all kinds of imaging equipment, which may provide important prior information about the modulation transfer function (MTF), i.e., the frequency representation of the blur, such as the possible MTF models with unknown parameters. Based on this prior, we aim to identify the actual degradation MTF models and estimate their corresponding parameters for image deblurring. We firstly analyse the statistical character of the blurred image from the fractal model; and then estimate the noise level by the high frequency energy of the blurred image. For given MTF degradation model, we estimate the fractal parameters and the MTF parameters by Maximum Likelihood Estimation (MLE), which is achieved numerically by alternating optimization scheme. For unknown MTF models, we propose likelihood ratio test and joint parameter estimation for the identification of the single MTF model and multi-MTF models, respectively. Therefore, the degradation MTF models can be obtained before the image deconvolution, which means that the proposed method translates the blind image deconvolution to the non-blind one. The numerical experiments test the accuracy of the parameter estimation and the correctness of the MTF model identification.