Information Processing & Algorithm Laboratory

While most image deblurring techniques directly restore the spatial image variable, we propose an amplitude and phase decomposition recognizing the importance of accurate phase estimation in recovering sharp image details. To that end, we first develop novel linear minimum mean squared (LMMSE) estimators of the amplitude and phase of the blurred, noisy image observation. An iterative optimization algorithm follows that recovers the sharp image using the aforementioned LMMSE estimators. Finally, matrix parameters that are statistically determined and fixed in the iterative algorithm are now learned using a training dataset of clean and degraded observations. Our deblurring engine is dubbed UPADNet (Unrolled Phase and Amplitude Decomposition Network), such that each iteration of the underlying phase and amplitude recovery algorithm is parameterized and trained end-to-end. Experiments over benchmark evaluation datasets such as GoPro, RealBlur and COCO datasets confirm that UPADNet outperforms state of the art deep networks including those based on algorithm unrolling in the image domain. The benefits of UPADNet are even more pronounced in high noise and limited training data regimes.

This paper makes three primary contributions to blind image deblurring. First, it challenges the conventional practice of treating frequency-domain information implicitly by explicitly modeling the phase and amplitude components of blurred images, highlighting the critical role of phase in preserving structural details during restoration. Second, it introduces a statistically grounded formulation by deriving Linear Minimum Mean Squared Error (LMMSE) estimators for both amplitude and phase, providing a principled alternative to conventional Fourier-domain approximations and demonstrating improved estimation accuracy under noise. Third, the proposed optimization framework is transformed into UPADNet, a model-driven deep unrolled architecture that learns the parameters of the iterative reconstruction process directly from data while preserving interpretability. The resulting framework effectively combines physical modeling, statistical estimation, and deep learning, achieving state-of-the-art deblurring performance with strong robustness to noise, real-world degradations, and limited training data, demonstrating improved generalization compared with existing deep and unrolled methods.

S. Malek et al., "Leveraging Phase Information to Boost Unrolled Network Learning for Image Deblurring". [ECCV2026]

UPADNet

Unrolled Phase and Amplitude Decomposition Network

European Conference on Computer Vision (ECCV)

ABSTRACT

Key Contributions

UPADNet

Results

Results in nosiy condition

Published Paper