Towards Differential Handling of Various Blur Regions for Accurate Image Deblurring

TL;DR

This work tackles image deblurring when blur degradation varies across regions by introducing DHNet, a differential handling network that couples a Volterra-based Volterra Block (VBlock) with a Degradation Degree Recognition Expert Module (DDRE). The VBlock models nonlinearities with higher-order interactions via Volterra kernels, avoiding heavy stacking of activation functions, while DDRE leverages prior knowledge to estimate spatially varying blur and routes information through multiple experts with adaptive weights. Comprehensive GoPro, HIDE, and RealBlur experiments demonstrate state-of-the-art PSNR/SSIM with reduced computation, and extensive ablations validate the contributions of both VBlock and DDRE. The approach yields accurate deblurring across diverse blur regions, offering a practical and scalable solution for real-world applications.

Abstract

Image deblurring aims to restore high-quality images by removing undesired degradation. Although existing methods have yielded promising results, they either overlook the varying degrees of degradation across different regions of the blurred image, or they approximate nonlinear function properties by stacking numerous nonlinear activation functions. In this paper, we propose a differential handling network (DHNet) to perform differential processing for different blur regions. Specifically, we design a Volterra block (VBlock) to integrate the nonlinear characteristics into the deblurring network, avoiding the previous operation of stacking the number of nonlinear activation functions to map complex input-output relationships. To enable the model to adaptively address varying degradation degrees in blurred regions, we devise the degradation degree recognition expert module (DDRE). This module initially incorporates prior knowledge from a well-trained model to estimate spatially variable blur information. Consequently, the router can map the learned degradation representation and allocate weights to experts according to both the degree of degradation and the size of the regions. Comprehensive experimental results show that DHNet effectively surpasses state-of-the-art (SOTA) methods on both synthetic and real-world datasets.

Figures (13)

• Figure 1: Varying degrees of degradation across different regions. The top row is the blurred image and the bottom row is the sharp image.
• Figure 2: Computational cost vs. PSNR of models on the GoPro dataset Gopro. Our DHNet achieve the SOTA performance with up to 48.9% of cost reduction.
• Figure 3: The overall architecture of the proposed differential handling network (DHNet) for image deblurring mainly consists of the differential handling block (DHBlock), which includes the Volterra block (VBlock) and the degradation degree recognition expert module (DDRE). DDRE is shown in the one-router case for clarity.
• Figure 4: Image motion deblurring comparisons on the GoPro dataset Gopro. Our DHNet recovers image with clearer details.
• Figure 5: Image motion deblurring comparisons on the RealBlur dataset realblurrim_2020_ECCV. Our DHNet recovers perceptually faithful images.