No reference perceptual quality metrics: approaches and limitations

No-reference image and video quality estimation: Applications and human-motivated design

“…Blockiness, blurriness, and ringing features have also been combined with other features, including bitstream features [128,129] and edge gradient and luminance masking features [130].…”

The Consumer Digital Video Library [Best of the Web]

“…In general, industry wants an objective video-quality model that uses only the impaired video to predict MOS [4]. This approach is called no-reference (NR) and could be deployed on any video stream.…”

“…Based on the VQEG model validation reports, the ITU approved recommendations that include FR and RR models. By contrast, NR models have shown poor performance in VQEG validation testing [4], and no ITU recommendations were approved. This absence has stimulated interest in bit stream models and hybrid perceptual bit stream (hybrid) models.…”

Quality estimation based on interest points through hierarchical saliency maps

“…At the opposite of FR metrics, NR metrics aim at evaluating distorted images without any cue from the source. However, most of the proposed NR quality metrics [6] are designed for one or a set of specific distortions and are unlikely to be generalized for evaluating images with other types of distortion. While RR [7] metrics are between FR and NR, they make use of a part of the information from original images in order to evaluate the visual quality of the distorted ones.…”