Study On Blind Deconvolution Of Noisy Images

Method of image restoration includes the Inverse Filter, Wiener Filter, Constrained Least Squares Filter and so on. Blind Image Deconvolution is a focus in image restoration field nowadays. Improved constrained least squares under the condition of noised image and the approach to realize blur identification of blurred image as well as automatically image restoration are mainly discussed in the thesis. Automatically image restoration is referred to realize blur identification and estimation of normalization parameter and image restoration under the condition of giving blurred image and no interruption, the methods decrease the parameter inaccurate of interrupt and improve the restoration quality.At present, the literature has made a study on blur identification. The thesis improves the ARMA parameter estimation method, introducing smooth limit, and enhances the accurate of blur identification under the condition of noised image and realizes the automation of blur identification. With the development of every subject and the application of technology, all kinds of math tools are applied to image restoration, but constrained least square filter is still widely used as a traditional method because of its concision, reasonableness, computation speed, real-time ability and good noise suppression. It is proposed in the thesis that GCV method be applied to estimate the normalization parameter of Constrained Least Squares Filter and automation of image restoration is realized.The thesis has made a study on blur identification of noised image and image restoration, and GCV is applied to blur identification and estimation of normalization parameter of Constrained Least Squares Filter. It is proved in the experiment that image restoration automation is realized and the chance of interrupt is reduced. The thesis summarizes the advantage and disadvantage of every method and proposes several developing orientation in future study.