INDIGO+: A Unified INN-Guided Probabilistic Diffusion Algorithm for Blind and Non-Blind Image Restoration

TL;DR

INDIGO+ introduces an INN-guided probabilistic diffusion framework that unifies non-blind and blind image restoration by leveraging an invertible neural network to simulate and invert degradation during diffusion sampling. The forward INN enforces data consistency with measurements, while the inverse INN preserves diffusion-driven texture details, guided by gradient steps from the consistency-imposed intermediate reconstructions. The non-blind variant uses a fixed degradation simulator, whereas the blind variant adapts to unknown degradations through a conditional INN controlled by a degradation embedding learned from data and refined during testing. Across synthetic and real-world degradations, INDIGO+ achieves competitive or superior results compared with state-of-the-art diffusion and supervised methods, while offering initialization-based acceleration and the option to finetune for unseen degradations. This approach highlights the practical impact of combining invertible priors with powerful generative diffusion models for robust, flexible IR in real-world imaging pipelines.

Abstract

Generative diffusion models are becoming one of the most popular prior in image restoration (IR) tasks due to their remarkable ability to generate realistic natural images. Despite achieving satisfactory results, IR methods based on diffusion models present several limitations. First of all, most non-blind approaches require an analytical expression of the degradation model to guide the sampling process. Secondly, most existing blind approaches rely on families of pre-defined degradation models for training their deep networks. The above issues limit the flexibility of these approaches and so their ability to handle real-world degradation tasks. In this paper, we propose a novel INN-guided probabilistic diffusion algorithm for non-blind and blind image restoration, namely INDIGO and BlindINDIGO, which combines the merits of the perfect reconstruction property of invertible neural networks (INN) with the strong generative capabilities of pre-trained diffusion models. Specifically, we train the forward process of the INN to simulate an arbitrary degradation process and use the inverse to obtain an intermediate image that we use to guide the reverse diffusion sampling process through a gradient step. We also introduce an initialization strategy, to further improve the performance and inference speed of our algorithm. Experiments demonstrate that our algorithm obtains competitive results compared with recently leading methods both quantitatively and visually on synthetic and real-world low-quality images.

Figures (19)

• Figure 1: Comparisons with state-of-the-art blind image restoration approaches wang2023dr2yue2022diffaceyang2023pgdiffstablesr on the real-world low-quality images. Our algorithm produces high-quality reconstruction results and preserves more details than the recent leading methods. (Zoom in for best view).
• Figure 2: The wavelet transform obtained using the lifting scheme.
• Figure 3: Overview of our INDIGO for non-blind image restoration. Given a degraded image $\bm{y}$ during inference, the diffusion posterior sampling is guided by our data-consistency step with INN at each step $t$. We show the detailed algorithm in Algorithm \ref{['algo:nonblind']}.
• Figure 4: Overview of our BlindINDIGO for blind image restoration. Given a degraded image $\bm{y}$ during inference, our approach first predicts a clean version $\bm{y}_{0}$ with the Initialization Prediction Network (IPN) and extract an implicit degradation embedding $\gamma_{deg}$ with the Degradation Estimation Module (DEM). Next, starting from a diffused $\bm{y}_{0}$, the diffusion posterior sampling is guided by our data-consistency step with INN at each step $t$. We show the detailed algorithm in Algorithm \ref{['algo:blind']}.
• Figure 5: The forward and inverse transform of our INN during inference.