MCLEFusion: A multi-level correction low-light enhancement method for nighttime infrared and visible image fusion

To address the challenge of feature extraction caused by degraded detail and texture in nighttime visible images for optical sensing applications, this paper presents MCLEFusion, a method that integrates unsupervised low-light enhancement and infrared-visible fusion. Specifically, a Retinex-based multi-level low-light enhancement block (LEB) utilizes stage-by-stage enhancement, calibration, and targeted loss functions to significantly improve brightness and detail in low-light visible images while mitigating color distortion and over-smoothing. For fusion, a multi-feature residual block (MRB) employs gradient residual flow and multi-level feature extraction to efficiently fuse infrared and enhanced visible images. Extensive evaluations on public datasets confirm MCLEFusion's superior performance over existing methods in color consistency, detail preservation, and overall fusion quality, thereby improving information extraction from optical sensor data.