The field of image fusion and multimodal image processing is witnessing significant advancements, particularly in the areas of infrared-visible image fusion, multimodal image matching, high dynamic range imaging, and person re-identification across different modalities. A common theme across recent research is the emphasis on overcoming challenges related to data compatibility, modality invariance, and the efficient integration of diverse imaging techniques to enhance visual performance and task adaptability.

Innovative approaches are being developed to address the limitations of existing methods, such as the struggle with multimodal data compatibility and the generation of high-quality images from scenes with extreme dynamic ranges. Techniques leveraging deep learning and generative models are at the forefront, offering solutions that improve robustness, generalization, and the quality of fused images. Additionally, there is a growing focus on the security and robustness of multimodal person re-identification systems, with novel adversarial attack methods being proposed to test and improve the resilience of these systems.

Noteworthy papers include:

• A comprehensive survey on infrared-visible image fusion, providing a multi-dimensional framework and detailed analysis of current methods.
• MIFNet, which introduces a novel approach to learning modality-invariant features for multimodal image matching without the need for aligned multimodal training data.
• UltraFusion, a groundbreaking exposure fusion technique capable of handling images with up to 9 stops difference, setting a new standard for high dynamic range imaging.
• A novel Modality Unified Attack method for omni-modality person re-identification models, demonstrating significant advancements in the robustness of surveillance systems.
• The introduction of a mixed-modal ReID setting and the MixER method, which significantly improves performance in mixed gallery applications by disentangling modality-specific and shared identity information.