F2former: When Fractional Fourier Meets Deep Wiener Deconvolution and
Selective Frequency Transformer for Image Deblurring
1Space Applications Centre (SAC), 2Indian Institute of Space Science and Technology (IIST)
Proposed Fractional Fourier based Transformer (F2former) leverages the property of better handling of non-stationary signals
like images for different image deblurring scenarios. As fractional Fourier transform (FRFT) analyses the joint distribution
between spatial and frequency domain information, our FRFT based model outperforms other SOTA methods especially in complex
scenarios like non-uniform and real-world blur situations as shown in above test scenarios for different datasets.
Abstract
Recent progress in image deblurring techniques focuses mainly on operating in both frequency and spatial domains using the Fourier
transform (FT) properties. However, their performance is limited due to the dependency of FT on stationary signals and its lack of
capability to extract spatial-frequency properties. In this paper, we propose a novel approach based on the Fractional Fourier Transform
(FRFT), a unified spatial-frequency representation leveraging both spatial and frequency components simultaneously, making it ideal for
processing non-stationary signals like images. Specifically, we introduce a Fractional Fourier Transformer (F2former), where we combine
the classical fractional Fourier based Wiener deconvolution (F2WD) as well as a multi-branch encoder-decoder transformer based on a new
fractional frequency aware transformer block (F2TB). We design F2TB consisting of a fractional frequency aware self-attention (F2SA) to
estimate element-wise product attention based on important frequency components and a novel feed-forward network based on frequency
division multiplexing (FM-FFN) to refine high and low frequency features separately for efficient latent clear image restoration.
Experimental results for the cases of both motion deblurring as well as defocus deblurring show that the performance of our proposed
method is superior to other state-of-the-art (SOTA) approaches.
Contributions
1. Design of a Fractional Feature-based Wiener Deconvolution (F2WD) layer to enhance the shallow-level features given a blurry input. As FRFT is capable of capturing spatially varying artefacts, F2WD efficiently performs deblurring in feature space.
2. We develop a novel building block, Fractional Hybrid Transformer Block (FHTB) to efficiently reconstruct the deblurred image. FHTB is composed of Fractional Fourier based Feature Refinement module (F3RB) and also a Transformer Block (F2TB) to extract local and global context features, respectively.
3. We design F2TB with Fractional Frequency aware Self-Attention (F2SA) for efficient computation and selective emphasis on key frequency components, and Frequency Division Multiplexing-based FFN (FM-FFN) to dynamically extract high and low frequency features using a cosine bell function for optimal frequency recovery.
4. We conduct experiments to demonstrate the effectiveness of the proposed F2former for motion and defocus blurring, showing significant performance improvements over other SOTA models. We also provide a detailed ablation study to validate the contribution of each module.
Bibtex
@article{paul2025f2former,
title={F2former: When Fractional Fourier Meets Deep Wiener Deconvolution and Selective Frequency Transformer for Image Deblurring},
author={Paul, Subhajit and Kumawat, Sahil and Gupta, Ashutosh and Mishra, Deepak},
journal={IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
year={2025}
}
