Image decomposition-based blind image deconvolution model by employing sparse representation

摘要

Conventional blind restoration methods often take consideration of images as a whole. However, an image may have different types of components, and each component has different morphology and properties. Using one model only can capture one part of images effectively, but fail to represent the others; the processing results by using conventional methods would lose some important features. In this study, a new sparse prior-based blind image deconvolution model has been proposed by employing commonly considered image decomposition strategy which separates an image into cartoon (piecewise-smooth part) and texture (the oscillating pattern part). On the basis of the different properties of cartoon and texture, it, respectively, regularises the texture with the sparsity of discrete cosine transform domain, and the cartoon with a combined term including framelet-domain-based sparse prior and a quadratic regularisation. Then a double alternating split Bregman iteration is proposed to address the proposed minimisation problem. It has been demonstrated that images can be recovered with high quality and more abundant features by authors' proposed algorithm than other popular deblurring methods.