The adversarial robustness of image quality assessment (IQA) models to adversarial attacks is emerging as a critical issue. Adversarial training has been widely used to improve the robustness of neural networks to adversarial attacks, but little in-depth research has examined adversarial training as a way to improve IQA model robustness. This study introduces an enhanced adversarial training approach tailored to IQA models; it adjusts the perceptual quality scores of adversarial images during training to enhance the correlation between an IQA model’s quality and the subjective quality scores. We also propose a new method for comparing IQA model robustness by measuring the Integral Robustness Score; this method evaluates the IQA model resistance to a set of adversarial perturbations with different magnitudes. We used our adversarial training approach to increase the robustness of five IQA models. Additionally, we tested the robustness of adversarially trained IQA models to 16 adversarial attacks and conducted an empirical probabilistic estimation of this feature.
1 January 2022 in Proceedings of the Institute for System Programming of the RAS
Our paper compares the accuracy of the vanilla ResNet-18 model with the accuracy of the Clipped BagNet-33 and BagNet-33 models with adversarial learning under different conditions. We performed experiments on images attacked by the adversarial sticker under conditions of image transformations. The adversarial sticker is a small region of the attacked image, inside which the pixel values can be changed indefinitely, and this can generate errors in the model prediction. The transformations of the attacked images in this paper simulate the distortions that appear in the physical world when a change in perspective, scale or lighting changes the image. Our experiments show that models from the BagNet family perform poorly on images in low quality. We also analyzed the effects of different types of transformations on the models' robustness to adversarial attacks and the tolerance of these attacks.
The adversarial robustness of image quality assessment (IQA) models to adversarial attacks is emerging as a critical issue. Adversarial training has been widely used to improve the robustness of neural networks to adversarial attacks, but little in-depth research has examined adversarial training as a way to improve IQA model robustness. This study introduces an enhanced adversarial training approach tailored to IQA models; it adjusts the perceptual quality scores of adversarial images during training to enhance the correlation between an IQA model’s quality and the subjective quality scores. We also propose a new method for comparing IQA model robustness by measuring the Integral Robustness Score; this method evaluates the IQA model resistance to a set of adversarial perturbations with different magnitudes. We used our adversarial training approach to increase the robustness of five IQA models. Additionally, we tested the robustness of adversarially trained IQA models to 16 adversarial attacks and conducted an empirical probabilistic estimation of this feature. The code is available at https://github.com/wianluna/metrics_at.
Recently, the area of adversarial attacks on image quality metrics has begun to be explored, whereas the area of defences remains under-researched. In this study, we aim to cover that case and check the transferability of adversarial purification defences from image classifiers to IQA methods. In this paper, we apply several widespread attacks on IQA models and examine the success of the defences against them. The purification methodologies covered different preprocessing techniques, including geometrical transformations, compression, denoising, and modern neural network-based methods. Also, we address the challenge of assessing the efficacy of a defensive methodology by proposing ways to estimate output visual quality and the success of neutralizing attacks. Defences were tested against attack on three IQA metrics -- Linearity, MetaIQA and SPAQ. The code for attacks and defences is available at: (link is hidden for a blind review).
Image Quality Assessment (IQA) models are increasingly relied upon to evaluate image quality in real-world systems -- from compression and enhancement to generation and streaming. Yet their adoption brings a fundamental risk: these models are inherently unstable. Adversarial manipulations can easily fool them, inflating scores and undermining trust. Traditionally, such vulnerabilities are addressed through data-driven defenses -- adversarial retraining, regularization, or input purification. But what if this is the wrong lens? What if robustness in perceptual models is not something to learn but something to design? In this work, we propose a provocative idea: robustness as an architectural prior. Rather than training models to resist perturbations, we reshape their internal structure to suppress sensitivity from the ground up. We achieve this by enforcing orthogonal information flow, constraining the network to norm-preserving operations -- and further stabilizing the system through pruning and fine-tuning. The result is a robust IQA architecture that withstands adversarial attacks without requiring adversarial training or significant changes to the original model. This approach suggests a shift in perspective: from optimizing robustness through data to engineering it through design.