Fidelity-Naturalness Evaluation of Single Image Super Resolution

TL;DR

This paper introduces a combined fidelity and naturalness evaluation framework for single image super resolution, improving upon traditional methods by incorporating human perception and bias correction.

Contribution

It proposes a new fidelity metric addressing bias issues and uses human pairwise comparisons to assess naturalness, enhancing super resolution evaluation.

Findings

Fidelity-naturalness method outperforms traditional evaluation techniques

New fidelity metric reduces bias in quality assessment

Human preferences correlate well with proposed naturalness metric

Abstract

We study the problem of evaluating super resolution methods. Traditional evaluation methods usually judge the quality of super resolved images based on a single measure of their difference with the original high resolution images. In this paper, we proposed to use both fidelity (the difference with original images) and naturalness (human visual perception of super resolved images) for evaluation. For fidelity evaluation, a new metric is proposed to solve the bias problem of traditional evaluation. For naturalness evaluation, we let humans label preference of super resolution results using pair-wise comparison, and test the correlation between human labeling results and image quality assessment metrics' outputs. Experimental results show that our fidelity-naturalness method is better than the traditional evaluation method for super resolution methods, which could help future research on single-image super resolution.