Spectral characteristics preserving image fusion based on Fourier domain filtering

Abstract

Data fusion methods are usually classified into three levels: pixel level (ikonic), feature level (symbolic) and knowledge or decision level. Here, we will focus on the development of ikonic techniques for image fusion. Image transforms such as the Intensity-Hue-Saturation (IHS) or Principal Component (PC) transform are widely used to fuse panchromatic images of high spatial resolution with multispectral images of lower resolution. These techniques create multispectral images of higher spatial resolution but usually at the cost that these transforms do not preserve the original color or spectral characteristics of the input image data. In this study, a new method for image fusion will be presented that is based on filtering in the Fourier domain. This method preserves the spectral characteristics of the lower resolution mul-tispectral images. Examples are presented for SPOT and Ikonos panchromatic images fused with Landsat TM and Iko-nos multispectral data. Comparison with existing fusion techniques such as IHS, PC or Brovey transform prove the su-periority of the new method. While in principle based on the IHS transform (which usually only works for three bands), the method is extended to any arbitrary number of spectral bands. Using this approach, this method can be applied to sharpen hyperspectral images without changing their spectral behavior.