PCTLFusion: A Progressive Fusion Network via Contextual Texture Learning for Infrared and Visible Image

The purpose of infrared and visible image fusion is to generate a fused image that contains rich texture details and highlights key targets. However, existing methods often result in sparse textures and poor visual quality under degraded lighting conditions. Additionally, most current fusion approaches focus only on the interaction of similar information during the encoding stage, while image fusion tasks require not only an emphasis on similarity but also on complementarity. In this work, we propose a progressive fusion network via contextual texture learning for infrared and visible image, termed PCTLFusion. To explicitly model texture information and enhance the limited texture in nighttime scenes, we design a Content and Context Texture Learning Module (CCTLM). Furthermore, we introduce an Affine Transformation-based Multimodal Pre-fusion Module (ATMPM) to learn and pre-fuse complementary and similar information, enabling multimodal information interaction during the encoding process. Through three iterations of CCTLM and ATMPM during encoding, multi-level features are effectively captured. Finally, the image is reconstructed using a decoder. Extensive experiments demonstrate that our proposed fusion algorithm outperforms state-of-the-art methods in both subjective and objective evaluations.