An Elliptical Basis Function Network for Non-blind Image Deconvolution

Biomedical imaging deblurring attempts to recover original human organ boundaries degraded by an imaging system due to low intensity signals and blur like defocus or motion blur that are frequently occurred. In this paper, we aim to deliver an improved procedure in deblurring images that are degraded, particularly CT medical images. Our work relies on probabilistic image patches prior Expected patch log likelihood (EPLL). The challenge is to find suitable prior in the presence of blur and noise. We propose a framework based on Gaussian Mixture prior trained by elliptical basis function (EBF) network to restore CT images. An expectation-maximization (EM) algorithm is used to estimate the basis functions. Extensive experimental results indicate in terms of visual quality, PSNR values that the EM-based EBF network outperforms the original EPLL-GMM model. The performance of the proposed method is assessed by experimental results pertaining to restoration of blurred images.