Regularization super-resolution image fusion considering inaccurate image registration and observation noise

Abstract

In this paper, a kind of super-resolution image fusion algorithm is proposed to regularize the distortion of the reconstructed high-resolution (HR) image caused by the inaccurate image registration and the observation noise. For this purpose, the registration error, caused by inaccurate image registration, is considered as the noise mean added in the observation noise known as additive white Gaussian noise (AWGN). Based on this consideration, two constraints are regulated pixel by pixel within the framework of Millerpsilas regularization, and combined with regularization parameters to construct one cost function. The regularization parameters are adaptively estimated in each pixel in terms of the registration error, as well as in each observation channel in terms of the AWGN. Simulation shows that the proposed regularized SR algorithm can fuse the information from multiple LR images effectively and achieve the reconstructed HR images with much sharper edges and higher PSNR.