New network improves ultra high definition low light images
Bidirectional Cross-frequency Domain Interaction Network for UHD Low-Light Image Enhancement
Computer Vision and Pattern Recognition
Summary
Low-light photos, especially in ultra-high definition, often look noisy, unevenly lit, or blurry. The authors created a new method called BiFreqNet that helps fix these photos by using both the light-and-dark areas and the fine details together. Their approach uses a special system to share information between different kinds of image features, making dark areas brighter and details clearer without adding noise. Tests show their method often works better than existing ones on several image sets.
What this means in practice
- •For photography software developers: Build tools that enhance ultra-high definition low-light photos by improving brightness and detail while reducing noise through dual-domain feature interactions.
- •For smartphone camera engineers: Integrate advanced low-light enhancement techniques to produce clearer, more detailed UHD images in challenging lighting conditions.$Commercial implications: Enables smartphone makers to offer superior low-light photo capture, enhancing user experience and device competitiveness.
Authors
Bo Fu, Hai Chi, Liyan Wang, Shilin Fu, Wei Wang, Kin-Man Lam, Jinshan Pan, Cong Wang
Abstract
Existing Ultra-High-Definition (UHD) low-light image enhancement (LLIE) methods typically rely on either spatial- or frequency-domain modeling, or independently process different frequency components for efficient restoration. However, they often fail to jointly exploit complementary spatial--frequency information and the inherent relationship between illumination and structural information across frequencies, leading to inconsistent illumination, residual noise, and detail artifacts. In this paper, we propose a novel Bidirectional Cross-frequency Domain Interaction Network (BiFreqNet) for UHD LLIE. Specifically, BiFreqNet incorporates a Dual-domain Feature Interaction (DFI) module into the multi-scale representations of a pretrained variational autoencoder (VAE) to perform reciprocal enhancement between low- and high-frequency features. DFI consists of a Spatial and Spectral-Aware Low-Frequency Enhancement (S$^2$LFE) block and a Cross-Frequency Enhancement (CFE) block. S$^2$LFE exploits high-frequency structural cues to guide low-frequency restoration and jointly models spatial and spectral information to recover global exposure and spatially varying illumination while suppressing sensor noise. CFE further leverages the enhanced low-frequency contextual information to guide adaptive high-frequency refinement, enabling faithful reconstruction of structures and fine details. Extensive experiments on four general and UHD LLIE benchmarks demonstrate that BiFreqNet achieves competitive or superior performance compared with state-of-the-art methods.