Adaptive Super-Resolution Image Reconstruction with Lorentzian Error Norm

Objectives: To focus on an inverse problem of reconstructing a high resolution image from set of captured low resolution (LR) frames. Methods/Statistical Analysis: The captured LR images are blurred, warped, down-sampled, noisy, and contains complementary information. Super resolution reconstruction(SRR) is a computational technique to correct the degradation that the captured images normally suffer and this problem is ill-posed due to blur and noise present in the captured frames, and regularization is imperative to obtain a stable solution. Findings: The proposed approach is based on a maximum-a-posteriori (MAP) framework by minimizing a cost function. Persuaded by the performance of Lorentzian norm in reducing the outliers and regularization parameter (λ) is obtained based on U-curve method, which significantly reduces the search interval, decreases the computation time, and step size is (β) is calculated using successive over relaxation (SOR) technique. Application/Improvements: SRR problem is solved by locating search interval for optimal λ based on the U-curve method and demonstrated in test/colour images, and frames extracted from a video.