Pixel Perfect: Relational Image Quality Assessment with Spatially-Aware Distortions
2026-05-04 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors propose a new way to judge image quality by comparing images instead of trying to give each image a fixed score. They create fake distorted images to train their model without needing people to label them. Their system can spot what kind of distortion an image has, how strong it is, and in what direction it differs from a reference image. Then, they use these insights to rank images by quality in a more detailed and understandable way than before. This method helps improve image processing without relying on costly human ratings.
Image Quality Assessment (IQA)Mean Opinion Score (MOS)Self-Supervised LearningDistortion MapContrastive LearningOrdinal RankingSynthetic DistortionAnti-Symmetric ObjectiveSpatially-Aware FeaturesRelational Quality Score
Authors
Fadeel Sher Khan, Long N. Le, Abhinau K. Venkataramanan, Seok-Jun Lee, Hamid R. Sheikh
Abstract
Traditional image quality assessment (IQA) methods rely on mean opinion scores (MOS), which are resource-intensive to collect and fail to provide interpretable, localized feedback on specific image distortions. We overcome these limitations by shifting from absolute quality prediction to a relational and directional assessment. Our approach utilizes a self-supervised synthetic distortion engine to generate training data, eliminating the need for manual annotation. A distortion prediction network is trained with an anti-symmetric objective to produce spatially-aware, disentangled maps that identify the type, intensity, and direction of distortions relative to a reference image. Subsequently, a scoring network is trained via contrastive learning on ordinally ranked image sets to predict a relational quality score. Our method provides a more granular and interpretable approach to IQA for the targeted optimization of image processing algorithms without requiring any human-labeled quality scores.