Multi-band image fusion via perceptual framework and multiscale texture saliency

Abstract

Multi-band images provide valuable complementary information, play an important role in target detection and recognition in complex environments, and have become a key research direction. However, existing fusion methods rarely consider or adapt to more than two bands of images and are easily affected by external physical conditions (e.g., variations in sensor characteristics and environmental illumination). This paper proposes a multi-band image fusion method based on a perception framework and multiscale texture saliency. By introducing the perception framework and the human visual system (HVS) space, the source images are decomposed into detail, feature, and base layers according to the perception characteristics of the human eye for texture granularity. Gabor filters were used to obtain the saliency of the fine-grained textures in the detail layer, thereby selectively extracting detailed texture information. The saliency of the feature layer texture was calculated using a Hessian matrix. Fusion weights were then obtained based on the texture complexity at the current scale, allowing for the effective extraction of structural information from the source images. Finally, fusion was performed in the unified framework of the HVS space, and the fusion image was obtained through an inverse transformation. The experimental results indicate that the multiscale texture perceptual framework fusion (MSTPFF) method can effectively transfer textures of different scales in the source images to the fused image, thus preserving the unique details and structural texture information of the multiband images. This transfer aligns with the visual characteristics of the human eye and significantly enhances fusion quality.