Selective filter with adaptive size macroblock for super-resolution applications

Abstract

Super-Resolution (SR) is a set of techniques which objective is to increase and improve the resolution of an image or a video sequence. In this scope, one of the most used techniques is “fusion”, where High-Resolution (HR) images are constructed from several observed Low-Resolution (LR) images. In this paper, a fusion SR algorithm is enhanced introducing an intelligent selective filter which decides the best LR frames to be used in the process. Additionally, an adaptive Macro-Block (MB) size decision maker has been developed to specify an appropriate frame division into MBs. This not only improves the quality but also reduces the computational cost of the baseline algorithm, avoiding the incorporation of non-correlated data. It is also presented how this new algorithm performs well with typical SR applications, such as underwater imagery, surveillance video or remote sensing. The algorithm results are provided on a test environment to objectively compare the quality enhancement of images processed by bilinear interpolation and the two aforementioned methods: Baseline and Enhanced SR, presenting a quantitative comparison based on the PSNR (Peak Signal-to-Noise Ratio) and the SSIM (Structural SIMilarity index).