Multispectral image fusion by super pixel statistics

Multispectral image fusion is a fundamental problem of image processing and remote sensing. This problem is addressed by both classic and deep learning approaches. This paper is focused on the classic solutions that can work in real-time systems and introduces a new novel approach to this group of works. The proposed method carries out multispectral image fusion based on the content of the fused images. Furthermore, it relies on an analysis of the level of information of segmented superpixels in the fused inputs. Specifically, the proposed method addresses the task of visible color RGB to Near-Infrared (NIR) fusion. The RGB image captures the color of the scene while the NIR channel captures details and sees beyond haze and clouds. Since each channel senses different information of the scene, their multispectral fusion is challenging and interesting. Therefore, the proposed method is designed to produce a fusion that contains the relevant content of each spectra. The experiments of this manuscript show that the proposed method is visually informative with respect to other classic fusion methods. Moreover, it can be run fastly on embedded devices without heavy computation requirements.