A novel resolution enhancement established on an iterative regularized ML MSRR and high-frequency pre-determining SSRR for ultra-magnification rate

Abstract

In almost all circumstances, a digital image with high spatial resolution (so called HR) is often pleasing however it is challenging to obtain due to overpriced cost on equipment devices. Consequently, Super Resolution Reconstruction (SRR) algorithm, which espouses algebraic formulation to obtain HR with reasonable cost, has been one of the most compelling research fields in Computer Vision (CV) and digital image processing (DIP). For enlarging up to 16× spatial resolution, this paper proposes a novel resolution enhancement established on cooperation of MSRR (Multi-frame Super Resolution Reconstruction) method and SSRR (Single-frame Super Resolution Reconstruction) method. First, to maintain the edges while get rid of the noise, a set of perverted images with low spatial resolution (so called LR) are used to reconstruct the better quality image with 4× spatial resolution magnification rate by using the MSRR method established on an iterative Tukey's Biweight regularized ML (Maximum Likelihood) technique. Later, this 4× image is used to reconstruct the 16× spatial resolution image by using SSRR method established the high-spectrum pre-determining. The verification results of numerical measurement (in PSNR) and visual measurement illustrate that the proposed resolution enhancement is successful for enlarging up to 16× spatial resolution and getting rid of the noise.