基于改进DeblurGAN的可见光图像去运动模糊方法

打开文本图片集

关键词：去运动模糊；可见光图像;DeblurGAN;光流一致性损失;ASFF;轻量化;运动模糊补偿中图分类号：TP391.41 文献标志码：A DOI：10.3969/j. issn.1673-3819.2026.02.010

A method for deblurring motion blur in visible light images based on improved DeblurGAN

WANGBilin

（Henan Technical Institute，Zhengzhou 45OO42，China）

Abstract：Thispaperfocusesonthetask of motiondebluring invisiblelightimages byimprovingandfurtherlightening DeblurGAN，resulting inthemodelFaster-DeblurGAN.The model introduces FasterNetandASFFmodulesintothegenerator networkandincorporatesopticalflowconsistencyloss.Whilemaintaininghighdeblurringperformance，itsignificantlyenhancesthenetwork's inferencespeedandreduces themodel'sparametercount.Experimentalresults indicatethatthemethod outperforms the benchmark model DeblurGAN on the GoPro dataset，with PSNR and SSIM improving by 3.3% and 2.4% ， respectively，and the parameter count is reduced by 52.9% . Thisnot only improvesperformance but also effectively realizes thelightweightmodificationofthemodel.Additionally，tdemonstratesbeterdeblurringeffctsinersofsujetivevisual quality，proving the effectiveness of the model’s improvements.

Key words：motiondebluring;visiblelight images；DeblurGAN;optical flowconsistencyloss；ASFF;lightweight;motion blur compensation

运动模糊是由于相机与被摄物体之间的相对运动在图像上产生的模糊效果。（剩余9334字）