FS-IQA: Certified Feature Smoothing for Robust Image Quality Assessment

2025-08-07

AI Summary

To address the insufficient robustness of image quality assessment (IQA) models against input perturbations and their difficulty in simultaneously preserving visual fidelity, this paper proposes Feature-space Randomized Smoothing (FSS), a certified defense method. FSS is the first to adapt randomized smoothing to the feature space—introducing feature-level noise without altering the network architecture—and establishes a theoretical linkage between feature perturbations and input perturbations. By leveraging the maximum singular value of the Jacobian matrix, FSS enables efficient robustness certification via a single forward pass. The method is applicable to both full-reference and no-reference IQA models. Evaluation on two benchmark datasets demonstrates that FSS improves inference speed by 99.5% (uncertified) and 20.6% (certified) over state-of-the-art approaches, while boosting correlation with human subjective scores by up to 30.9%. It thus significantly enhances both model robustness and perceptual consistency.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Feature Construction/Reformulation
• Natural Language Processing: Safety and Robustness

Application Category

• Security and Privacy: Large-scale security measurements
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

We propose a novel certified defense method for Image Quality Assessment (IQA) models based on randomized smoothing with noise applied in the feature space rather than the input space. Unlike prior approaches that inject Gaussian noise directly into input images, often degrading visual quality, our method preserves image fidelity while providing robustness guarantees. To formally connect noise levels in the feature space with corresponding input-space perturbations, we analyze the maximum singular value of the backbone network's Jacobian. Our approach supports both full-reference (FR) and no-reference (NR) IQA models without requiring any architectural modifications, suitable for various scenarios. It is also computationally efficient, requiring a single backbone forward pass per image. Compared to previous methods, it reduces inference time by 99.5% without certification and by 20.6% when certification is applied. We validate our method with extensive experiments on two benchmark datasets, involving six widely-used FR and NR IQA models and comparisons against five state-of-the-art certified defenses. Our results demonstrate consistent improvements in correlation with subjective quality scores by up to 30.9%.

Problem

• Enhancing robustness of IQA models via feature-space noise smoothing
• Preserving image fidelity while providing certified robustness guarantees
• Improving computational efficiency and accuracy in quality assessment

Innovation

• Feature space smoothing for robust IQA
• Jacobian analysis for noise-perturbation connection
• Efficient single-pass backbone processing