Subjective Image Quality Assessment With Boosted Triplet Comparisons

In subjective full-reference image quality assessment, a reference is distorted at increasing distortion levels. The differences between perceptual qualities of the and its versions are evaluated, often using degradation category ratings (DCR). However, DCR has been criticized since rating categories on this ordinal scale might not be perceptually equidistant, observers may have different understandings categories. Pair comparisons (PC) images, followed by Thurstonian reconstruction values, overcomes these problems. addition, PC more sensitive than DCR, it can provide values in fractional, just noticeable difference (JND) units that express precise perceptional interpretation. Still, comparison images nearly same difficult. We introduce boosting techniques embedded general triplet (TC) increase sensitivity even more. Boosting amplifies artefacts enlarges their visual representation zooming, increases visibility distortions flickering effect, or combines some above. Experimental results show effectiveness boosted TC for seven types (color diffusion, jitter, high sharpen, JPEG 2000 compression, lens blur, motion multiplicative noise). For our study, we crowdsourced over 1.7 million responses to questions. give detailed analysis data terms reconstructions, accuracy, detection rates, gain. Generally, discriminatory power allows reduce number without sacrificing accuracy resulting relative values. Our technique paves way fine-grained datasets, allowing levels, yet with high-quality annotations. also details from annotated dataset, KonFiG-IQA, containing 10 source processed 7 12 30 uniformly spaced span 3 JND units.