| In the last decade the number of commercially available imaging devices grew substantially. Nowadays a good quality photograph, in addition to tremendous selection of commercially available digital cameras, can be taken by numerous hand-held devices such as cell phones, smart phones, music players, and laptops. There has been an increase in the number of medical imaging systems that are used to diagnose abnormalities and to protect patients' health. In addition, digital images are available from a wide variety of technical areas such as astronomical imaging, remote sensing, and optics, among others.;Digital images obtained by imaging devices are never perfect as the observed images are often degraded by a measuring noise or blurred by a relative movement between the imaging device and the scene. This exponential increase in the number of imaging devices resulted in the need for more advanced image restoration algorithms. In the classical image restoration algorithms the blurring function is assumed to be known. Blind image deconvolution refers to the much more difficult problem in which both the original image and the blur are estimated from a degraded noisy observation by using only a partial information about the imaging system. Numerous blind image deconvolution algorithms have been appeared in the literature; they differ on the particular degradation model, the image and blur model, and the algorithmic framework they adopt.;In this dissertation, a hierarchical Bayesian framework is utilized to develop novel blind image deconvolution algorithms. The main contributions of this work are: 1) providing an estimation (within a hierarchical Bayesian framework) of the unknown image, blur and the hyperparameters utilized to model the image, blur and observation, 2) providing a computationally efficient estimation of the unknown image and hyperparameters utilized to model the image and the observation, 3) frameworks and algorithms for partial blur removal, and 4) frameworks and algorithms for compressive blind image deconvolution.;Experimental results demonstrate that the proposed algorithms provide restoration results that are ,on the average, superior to existing state-of-the-art methods. |
