An effective reconstructed pyramid crosspoint fusion for multimodal infrared and visible images

Fusion for multimodal infrared and visible images refers to the process of combining detail from both infrared and visible images, which holds particular importance in enhancing scene detection and tracking within video surveillance systems. However, traditional approaches for infrared and visible images fusion often struggle with preserving edges, capturing modality specific information effectively, enhancing detail across multiple scales, and efficiently optimizing fusion parameters. The proposed effective Edge preserving Gradient domain Guided image Filter with Weighting Approach method addresses these challenges by preserving edges during image decomposition and utilizing Multi Head Convolutional Split Attention networks for modality specific feature extraction. Additionally, the Pyramid Crosspoint Fusion Transformer model enhances detail and preserves crucial features through pyramid-based reconstruction, ensuring high quality representations. Employing the Humboldt squid optimization algorithm efficiently optimizes fusion parameters, further enhancing image quality. The suggested framework exhibits outstanding performance, achieving high standard deviation, mutual information, and information entropy of 85.21 nm, 16.32 bits, and 7.95 bits, respectively. This fusion model enhances scene detection and tracking in video surveillance systems by effectively combining infrared and visible images, preserving edges and details. It optimizes fusion quality through innovative techniques and evaluation metrics.