Blind image super-resolution using joint interpolation-restoration scheme

This paper presents a novel blind super-resolution (SR) algorithm to enhance image resolution without the knowledge of the blurring functions. The primary motivation for developing the blind algorithm is due to the observation that most existing SR methods assume the blurs are known a priori, which may not be true in many real-world applications. In view of this, we formulate the SR problem into a new framework of joint interpolation-restoration scheme. The main contribution of this paper is the development of multichannel blind restoration to estimate the unknown blurring functions, and its integration into the SR scheme to render high-resolution (HR) images. The blurs are identified iteratively with the utilization of inter-frame blur similarity. Experimental results show that the new algorithm is effective in performing blind SR image reconstruction.