Zamzam-Fusion for dual-gain with NLM-CDDFuse for CMOS sensors using ATEF-DRPI metric

This paper presents an enhanced version of the correlation-driven dual-branch feature decomposition framework (CDDFuse) for fusing low- and high-exposure images captured by the G400BSI sensor. We introduce a novel neural long-term memory (NLM) module into the CDDFuse architecture to improve feature extraction by leveraging persistent global feature representations across image sequences. The proposed method effectively preserves dynamic range and structural details, and is evaluated using a new metric, the ATEF dynamic range preservation index (ATEF-DRPI). Experimental results on a G400BSI dataset demonstrate superior fusion quality, with ATEF-DRPI scores of 0.90, a 12.5% improvement over that of the baseline CDDFuse (0.80), indicating better detail retention in bright and dark regions. This work advances image fusion techniques for extreme lighting conditions, offering improved performance for downstream vision tasks.