CCFusion: A Novel Channel-Convolutional Neural Network for Infrared and Visible Image Fusion

In recent years, deep learning–based methods have made significant advancements in infrared and visible image fusion. These methods improved fusion quality by designing complicated network architectures and loss function, overlook the representational mechanisms of multimodal image fusion. This leads to insufficient interpretability and the loss of critical information such as thermal targets in infrared images or texture details in visible images. To address this issue, we introduce a novel infrared and visible image fusion method based on channel-convolutional neural network, named CCFusion.The framework consists of two stages: feature extraction and feature fusion. DenseNet is adopted to extract modality-specific features, while a cross-matching concatenation strategy is designed to enhance the expression of feature information. Additionally, an optimized multi-scale channel attention module is integrated with the channel-convolutional neural network to fuse the feature maps, enhancing the representation of salient information while reducing redundancy. Finally, the Channel-convolutional neural network utilizes the weighted feature maps to generate the fusion image. Experiments are conducted on two public datasets (RoadScene, TNO) and one complex-scene image fusion dataset (AVMS). The results demonstrate the effectiveness of CCFusion. The comparison results show that CCFusion outperforms existing state-of-the-art infrared and visible image fusion methods in both qualitative and quantitative evaluations.