Abstract:
In this paper, we consider the super-resolution (SR) problem in a real scenario involving CCTV videos. As high-resolution (HR) ground-truth is usually not available for s...Show MoreMetadata
Abstract:
In this paper, we consider the super-resolution (SR) problem in a real scenario involving CCTV videos. As high-resolution (HR) ground-truth is usually not available for such an application, we suggest using no reference image quality assessment metric (e.g. BRISQUE, in this work [1], [20]), to evaluate two example single-image SR approaches suitable for such a task. Notwithstanding the unavailability of ground-truth, we also demonstrate that such a no-reference quality metric, is also more reliable in evaluating SR methods than popular metrics such as PSNR, SSIM even if a separate standard data with HR-LR pair is used to evaluate SR approaches. Finally, some real examples for CCTV images are shown for comparison of SR approaches using BRISQUE.
Date of Conference: 01-02 December 2018
Date Added to IEEE Xplore: 27 May 2019
ISBN Information:
Print on Demand(PoD) ISSN: 2164-7011