Local Dynamic Filter Network for Low-Light Enhancement and Deblurring

Nanxin Huang, Yirui Wang, Lifang Yang, Xianglin Huang, Nenghuan Zhang

Abstract: Under specific conditions, capturing clear images that meet human vision requirements is challenging due to limitations in electronic devices and scene capture. Traditional low-light enhancement methods primarily focus on brightness enhancement, neglecting issues like image blurring. This paper presents the Local Dynamic Filter Network (LDFN), an encoder-decoder framework that effectively restores texture information in low-light images. The encoder extracts multi-scale features, serving as input for the filter generation network, which produces local dynamic filters for the decoder. These filters restore texture details and reduce blur in the decoder. Additionally, pixel shuffle is utilized as an upsampling module. Experimental results on real-world and synthesized datasets demonstrate the method’s advantages and robustness. By addressing texture restoration and blur removal, LDFN offers a promising approach for enhancing low-light image quality.

External IDs:dblp:conf/iftc/HuangWYHZ23