Super-resolution Quality Criterion (SRQC): a super-resolution image quality assessment metric

Abstract

Recently, image super-resolution has reinforced image resolution enhancement approaches in real-time and ensuring visual quality of super resolved images has evolved as a key research problem. Most quantitative benchmarks rely on full reference metrics which would work in the presence of a reference image. However, the unavailability of ground truth images in real world applications and the size constraints of low resolution and high resolution images often pose major challenges to such metrics. In order to address these problems, we present a super-resolution image quality index (SRQC –Super-resolution Quality Criterion), which can effectively quantify the efficiency and performance of image super-resolution algorithms. SRQC benchmark evaluates the quality score of a super resolved image according to the perceptual concepts of low-level spatial features in high sharpness space and curvelet based quality-aware features from focal energy bands, which can be used to capture the quality preservation of an SR image. The proposed metric is referenceless, the significance being that the assessment does not require ground-truth image. Explicitly, the SR image is assessed in the curvelet domain which is suitable for the no-reference super-resolution image quality assessment based on human perception. Experimental scores illustrate that the SRQC is more competent in modeling the features from curvelet transform, thus quantifying the quality score of the super resolved image and outperforming the formerly reported image quality assessment metrics.