Toward Generalized Image Quality Assessment: Relaxing the Perfect Reference Quality Assumption

2025-03-14

AI Summary

Traditional full-reference image quality assessment (FR-IQA) assumes the reference image is perceptually perfect—a premise increasingly invalid due to modern imaging limitations and generative enhancement techniques (e.g., diffusion models), which often produce outputs superior to the original reference. This work systematically relaxes this strong assumption for the first time. We introduce DiffIQA, the first large-scale benchmark for generative-enhanced images featuring human-annotated triple-wise quality comparisons. We further propose A-FINE, an adaptive fidelity-naturalness evaluation model that employs multi-scale feature alignment and a dual-branch adaptive fusion mechanism to dynamically balance fidelity and naturalness. Extensive experiments demonstrate that A-FINE significantly outperforms conventional FR-IQA methods on DiffIQA, SRIQA-Bench, and mainstream IQA datasets. Both the code and the DiffIQA dataset are publicly released.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Natural Language Processing: Generation
• Machine Learning: Deep Generative Models & Autoencoders

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets

Abstract

Full-reference image quality assessment (FR-IQA) generally assumes that reference images are of perfect quality. However, this assumption is flawed due to the sensor and optical limitations of modern imaging systems. Moreover, recent generative enhancement methods are capable of producing images of higher quality than their original. All of these challenge the effectiveness and applicability of current FR-IQA models. To relax the assumption of perfect reference image quality, we build a large-scale IQA database, namely DiffIQA, containing approximately 180,000 images generated by a diffusion-based image enhancer with adjustable hyper-parameters. Each image is annotated by human subjects as either worse, similar, or better quality compared to its reference. Building on this, we present a generalized FR-IQA model, namely Adaptive Fidelity-Naturalness Evaluator (A-FINE), to accurately assess and adaptively combine the fidelity and naturalness of a test image. A-FINE aligns well with standard FR-IQA when the reference image is much more natural than the test image. We demonstrate by extensive experiments that A-FINE surpasses standard FR-IQA models on well-established IQA datasets and our newly created DiffIQA. To further validate A-FINE, we additionally construct a super-resolution IQA benchmark (SRIQA-Bench), encompassing test images derived from ten state-of-the-art SR methods with reliable human quality annotations. Tests on SRIQA-Bench re-affirm the advantages of A-FINE. The code and dataset are available at https://tianhewu.github.io/A-FINE-page.github.io/.

Problem

• Challenges the assumption of perfect reference image quality in FR-IQA models.
• Proposes a generalized FR-IQA model to assess fidelity and naturalness adaptively.
• Validates the model using a new super-resolution IQA benchmark.

Innovation

• Developed DiffIQA database with 180,000 images
• Created A-FINE model for image quality assessment
• Introduced SRIQA-Bench for super-resolution validation