Perceptual loss functions in Deep Neural Network (DNN) deblurring architectures improve the overall quality of restored images. However, few focus on explicitly targeting sharpness in the restorations. We conduct a subjective study of models trained with and without losses which explicitly target sharpness using a four-protocol approach, exploring preferred sharpness levels and effects on image quality. We introduce a novel dataset of images with uniform sharpness increments along with Difference Mean Opinion Scores (DMOS). Additionally, we propose a novel class of Sharpness Informed (SI) Image Quality Assessment (IQA) metrics which properly penalize over-sharpening. Our new SI-PSNR metric outperforms all other PSNR variants in terms of correlation statistics on IQA benchmarking datasets. We show that, on average, images restored using a sharpness-aware composite loss are preferred in 67% of binarized comparisons, as opposed to losses that do not explicitly target sharpness.
Figures & tables
Fig. 1 : Ringing Map generated using Ringing Detection Ratio ( α ) with a patch size of 4 × 4 . Left to Right: GT image, Over-sharpened image, Ringing Detection Heatmap. Areas in red (mostly edges) correspond to heavy ringing which are penalized more than areas in blue (flat patches).
Fig. 2 : Sharpness ( Q ) Comparison of Reference Image and Sharpened Image . Images were sharpened to their preferred sharpness levels, showing that subjects preferred sharper images overall, as compared to the reference image.
Fig. 3 : Example of restorations from different DNN-based deblurring mechanisms with different losses . (Top, Bottom): Images restored using ARKNet [ 6 ] and XY-Deblur [ 15 ] . (Left, Right): Images restored using L and Lc . Images restored with Lc from both deblurring methods produce slightly sharper edges, resulting in aesthetically better images.
Fig. 4 : Comparison of MOS for images restored with separate losses averaged over two separate DNN models: ARKNet and XY-Deblur . Red dots higher than black dots denote that subjects preferred images restored with Lc rather than L , showing that subjects preferred slightly sharper restorations.
Fig. 5 : Example of an image restored with composite loss Lc (SI-PSNR = 27.26) and standard loss L (SI-PSNR = 35.15). Such images were shown to subjects within the framework of Protocol 2, wherein they were asked to select the sharper of the two.
Dataset
SSIM [ 3 ]
LPIPS [ 5 ]
VIF [ 22 ]
GSSIM [ 21 ]
Q [ 7 ]
BRISQUE [ 10 ]
NIMA [ 19 ]
PSNR
PSNR-HVS-M [ 4 ]
PSNR-HMA [ 20 ]
SI-PSNR
SI-SSIM
PLCC
0.885
0.921
0.791
0.909
0.462
0.568
0.135
0.876
0.873
0.875
0.905
0.886
KRCC
0.702
0.741
0.600
0.706
0.320
0.405
0.098
0.696
0.685
0.680
0.724
0.672
KADID-10K [ 18 ]
SROCC
0.887
0.915
0.794
0.893
0.461
0.582
0.119
0.877
0.871
0.866
0.901
0.871
PLCC
0.912
0.855
0.720
0.887
0.536
0.624
0.315
0.720
0.761
0.702
0.857
0.910
KRCC
0.753
0.686
0.545
0.712
0.381
0.432
0.146
0.554
0.589
0.540
0.681
0.758
Our Dataset
SROCC
0.920
0.870
0.728
0.900
0.519
0.612
0.163
0.740
0.783
0.726
0.869
0.923
TABLE I : Correlation between IQA metrics and DMOS over two separate datasets . The columns highlighted in grey are NR-IQA. The best and second best performing metrics are highlighted in orange and blue respectively. Our proposed metrics are in italics, bold and underlined (cf. last two columns). Our proposed SI-PSNR produces the best correlation in comparison to all other PSNR variants.
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.
Fadeel Sher Khan, Long N. Le, Abhinau K. Venkataramanan +2
The University of Texas at Austin Austin, TX, USA · Samsung Research America Plano, TX, USA
In this paper, we introduce a metric-guided dynamic loss controller (DLC) for multi-objective image restoration. Conventional image restoration pipelines usually train with a fixed weighted combination of multiple losses, without changing the relative importance of fidelity, perceptual similarity, and no-reference quality during optimization. DLC is an architecture- and loss-term-agnostic training-time controller: it does not modify the restoration architecture or introduce new differentiable loss terms, but dynamically reweights the existing training losses. During training, DLC periodically evaluates the current model on a small fixed feedback subset and uses the resulting quality metrics to update the loss-weight vector through an LLM-based controller. Because DLC operates on existing loss terms rather than task-specific architectures, the same controller formulation can be instantiated across diverse image restoration training pipelines. We evaluate DLC on three restoration domains: low-light image enhancement, deraining, and real-world super-resolution, using both reference-based and no-reference quality metrics. Across these settings, DLC considers metric-dependent trade-offs during optimization and guides training toward balanced operating points across fidelity and perceptual quality. The results show that DLC can move models toward more favorable operating points across different restoration domains, supporting its role as a practical plug-in controller for multi-objective image restoration.
Jaewan Ko, Janghoon Choi
Major in Data Science Convergence, Graduate School of Data Science Kyungpook National University, Daegu, Republic of Korea
Multi-modal large language models (MLLMs) have demonstrated significant potential in image quality assessment (IQA) by bridging visual perception with descriptive evaluations. However, existing approaches mainly focus on holistic quality prediction, often functioning as black boxes that provide limited insight into where distortions occur and how they affect perceived quality, hindering fine-grained analysis of localized and heterogeneous degradations. We propose GS-IQA, a framework that reformulates IQA as a progressive Where--What--How diagnosis, emulating the human perceptual process from an initial glance to closer scrutiny. Since a severity judgment is meaningful only for a correctly localized and recognized region, we realize this progression through a two-stage reinforcement learning paradigm that respects such dependencies: the glance stage uses a perception-gated reward to establish where degradations lie and what they are, activating severity feedback only once both are correct, while the scrutiny stage introduces online reward-conditioned degradation generation to synthesize hard examples targeted at the model's perceptual bottlenecks, sharpening its discrimination of subtle severity variations. To enable systematic evaluation, we construct Diag-Bench, a region-level IQA benchmark of about 25K curated samples spanning 12 distortion types and five ordinal severity levels. Extensive experiments show that GS-IQA consistently surpasses state-of-the-art methods in distortion localization, recognition, and severity estimation, and that its diagnostic representations transfer effectively to conventional global quality prediction across diverse external benchmarks. Code and data will be released.
Aoting Zhang, Mingze Gao, Dongbao Yang +5
IIE, Chinese Academy of Sciences · University of Chinese Academy of Sciences · Xiaohongshu Inc. +1