This paper proposes DPEC (Diffusion Prior-based Exposure Correction), a novel framework for image exposure correction that utilizes diffusion-based image generation priors encapsulated in pre-trained large-scale diffusion models, and seamlessly combines the strengths of diffusion models and regression models.
Abstract
Although most existing exposure correction methods achieve high fidelity, they often place excessive focus on overall pixel-wise accuracy, making it challenging to effectively model extreme exposure regions, which results in suboptimal perceptual quality. Recently, diffusion models have received significant attention due to their remarkable performance in the realm of image generation. However, their successful application to exposure correction remains a challenging and open question. The key challenge lies in generating accurate image structures and maintaining high image fidelity during stochastic diffusion processes. In this paper, we propose DPEC (Diffusion Prior-based Exposure Correction), a novel framework for image exposure correction that utilizes diffusion-based image generation priors encapsulated in pre-trained large-scale diffusion models. Specifically, we first propose an efficient fine-tuning strategy to derive an exposure corrector from pre-trained models, enabling the generation of enhanced images in a single-step denoising process. Moreover, we seamlessly combine the strengths of diffusion models and regression models, and design a joint cross-attention module to integrate multi-scale diffusion prior features, thereby effectively preserving high-frequency details and minimizing random artifacts. The diffusion model focuses on dealing with low-frequency content rather than all the intricate texture details. The experimental results demonstrate that the proposed DPEC method consistently outperforms existing state-of-the-art methods on multiple exposure correction datasets, whether in terms of fidelity, perceptual quality, or visual effects.
UGDiff, a novel diffusion guidance paradigm designed to further improve the perception-distortion balance, is introduced, which first estimates the reconstruction uncertainty of the latent features corresponding to a high-fidelity image and guides the diffusion process to selectively restore high-frequency details in high-uncertainty regions, while preserving fidelity elsewhere.
Ren Wang, Yung-Yu Chuang· International Conference on...· 0 citations
In recent years, diffusion models have emerged as a prominent research focus in the field of image restoration (IR) due to their demonstrated diversity and interpretability in generating images. IR aims to recover degraded images resulting from various conditions. However, the significant distributional discrepancies of different degraded images in the feature space present a challenge, making it difficult for existing diffusion models to simultaneously address the heterogeneity in restoration requirements across tasks. To overcome this limitation, we design a novel diffusion framework. This framework employs interactive modeling of determinism and stochasticity inherent in multiple image degradation processes. Consequently, the diffusion process effectively adapts to the specific demands of different tasks, whether prioritizing precision or diversity in the restored images. In contrast to previous deep learning-based IR methods, our approach provides stronger interpretability for the restoration process. Within this framework, the deterministic modeling component offers clear guidance for the reverse generative process of diffusion, ensuring restoration accuracy. Simultaneously, the stochastic modeling component enhances the model’s generalization capabilities under complex conditions, enabling it to handle diverse IR scenarios. We conduct experiments on four representative IR tasks: image deraining, image denoising, shadow removal, and low-light image enhancement. The proposed method achieves superior performance across all these tasks, demonstrating excellent generality and adaptability. Source code is available at https://github.com/AHU-psy/ILDS-mian.
Sha Luo, Si-Yu Peng, Da-Wei Zhao et al.· The Visual Computer· 0 citations
Observation Operator Diffusion is proposed, a unified framework that aligns both the supervision trajectory and feature refinement with the intrinsic recovery order of image structures and introduces GL-CoDA, a decoder that injects scale-specific Gaussian-Lanczos observations across decoding stages for coarse-to-fine feature refinement.
Shaojie Guo, Li-Chen Ma, Haoyang Tong et al.· 0 citations
Image super-resolution, which aims to reconstruct high-resolution images from their low-resolution observations, is fundamental to medical imaging, remote sensing, surveillance, microscopy, and scientific visualization. Traditional model-based methods formulate super-resolution as an inverse problem with hand-crafted regularization priors. While interpretable and theoretically grounded, they rely on fixed assumptions and require computationally intensive iterative solvers. Deep learning methods offer data-driven flexibility by learning nonlinear mappings from low- to high-resolution images, among which diffusion models have achieved particularly impressive perceptual quality. However, the standard diffusion training objective is a pixel-domain noise-prediction loss that does not explicitly enforce perceptual fidelity, which can lead to oversmoothing and loss of fine image structure. To address these limitations, we propose a perceptually regularized diffusion framework that incorporates prior knowledge through perceptual-loss-based regularization, improving training convergence and encouraging the recovery of meaningful image features. Experiments on benchmark datasets demonstrate improved perceptual quality and competitive distortion metrics, highlighting the effectiveness of regularization for diffusion-based super resolution.
Chuxiang Wang, Pavithra Venkatachalapathy, Ying Liang et al.· 0 citations
FSP-Diff, a novel one-step diffusion model featuring a dual-pathway architecture that refines semantic guidance using structured details to mitigate semantic deviations, is proposed and demonstrates that FSP-Diff surpasses existing one-step diffusion methods in both quantitative and qualitative metrics.
Chunxiao Liu, Wei Liu, Anbin Xiong et al.· 0 citations
To mitigate phototoxicity, photobleaching, and motion blur, long-term live-cell fluorescence imaging is typically performed with low excitation intensity and short exposure times, resulting in extremely low and progressively decaying signal-to-noise ratio (SNR). Meanwhile, the inherent irreproducibility of biological processes makes it exceedingly difficult to acquire ground-truth images. Under these conditions, existing denoising methods often suffer from structural detail loss. To address these challenges, we propose diffusion-guided unsupervised adaptive learning (DUAL), a two-stage self-supervised denoising framework that integrates regression-based and generative models. In the first stage, a regression module produces a stable structural estimate that preserves the underlying biological morphology. In the second stage, a conditional diffusion model restores high-frequency details under the constraint of this estimate, while effectively suppressing structural hallucinations. To further accommodate time-varying SNR, we introduce an adaptive diffusion inversion strategy that dynamically aligns the denoising strength with the degradation level of the input. In addition, temporal conditioning is incorporated to enhance cross-frame consistency. Extensive experiments on synthetic and real dynamic imaging datasets demonstrate that DUAL achieves superior denoising performance. It effectively improves noise suppression, structural preservation, and temporal consistency across diverse fluorescence imaging scenarios with low and time-varying SNR, providing a powerful computational solution for long-term fluorescence microscopy enhancement.
