Infrared image hierarchical enhancement network via guided decomposition

Infrared (IR) imaging technology plays a crucial role in diverse fields. However, the visual quality of current IR images remains unsatisfactory, frequently characterized by low contrast and lack of edge texture details. Moreover, extracting and enhancing detailed features without amplifying noise remains challenging. In this study, an IR image enhancement network is proposed by employing guided decomposition. The proposed network consists of an image decomposition network (IDN) and an image enhancement network (IEN). A guided convolutional module (GCM) based on guided filtering principle is introduced. Using this module, the IDN decomposes the input IR image into a base layer and a detail layer. The IEN enhances the detail layer and base layer separately via distinct network structures with corresponding loss functions. The enhanced representations of both layers are then fused through weighted parameters. Experimental results demonstrate that the proposed method outperforms the benchmark methods across both quantitative and qualitative evaluations.