2018/09/17 by Thomas Köhler, Köhler, Thomas, Michel Bätz +9
Biochemistry, Genetics and Molecular Biology · Computer Science · #Advanced Fluorescence Microscopy Techniques #Advanced Image Processing Techniques #Advanced Vision and Imaging #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Image Enhancement Techniques
paper · pdf · doi:10.48550/arxiv.1809.06420
openalex publication_date 2018/09/17 · openalex created_date 2022/08/03 · openalex updated_date 2026/07/28
Capturing ground truth data to benchmark super-resolution (SR) is\nchallenging. Therefore, current quantitative studies are mainly evaluated on\nsimulated data artificially sampled from ground truth images. We argue that\nsuch evaluations overestimate the actual performance of SR methods compared to\ntheir behavior on real images. Toward bridging this simulated-to-real gap, we\nintroduce the Super-Resolution Erlangen (SupER) database, the first\ncomprehensive laboratory SR database of all-real acquisitions with pixel-wise\nground truth. It consists of more than 80k images of 14 scenes combining\ndifferent facets: CMOS sensor noise, real sampling at four resolution levels,\nnine scene motion types, two photometric conditions, and lossy video coding at\nfive levels. As such, the database exceeds existing benchmarks by an order of\nmagnitude in quality and quantity. This paper also benchmarks 19 popular\nsingle-image and multi-frame algorithms on our data. The benchmark comprises a\nquantitative study by exploiting ground truth data and qualitative evaluations\nin a large-scale observer study. We also rigorously investigate agreements\nbetween both evaluations from a statistical perspective. One interesting result\nis that top-performing methods on simulated data may be surpassed by others on\nreal data. Our insights can spur further algorithm development, and the publicy\navailable dataset can foster future evaluations.\n