Influence of source images spatial characteristics on the global quality of fused images

Abstract

techniques to perform remote sensed image fusion are based on multiresolution analysis. This kind of images analysis requires the decomposition of the image at differente scales or levels, depending the fusion results on this level. Then, the two main objectives of this work are: to investigate the influence of the source images spatial characteristics on the decomposition level that the process fusion should be performed in; and to show how depends the spatial-spectral quality of fused images on this decomposition level. To carry out this study, the image fusion methodology that has been applied is based on the Wavelet transform, calculated by the à trous algorithm. The quality of the fused images has been evaluated by the ERGAS indices, as well as, the spectral correlation, the spatial correlation (Zhou’s index) and a global index (Q4). This methodology has been applied to fuse several multispectral and panchromatic images registered by the corresponding sensors on board the Landsat, Ikonos, and Quickbird satellites. It has been demonstrated that, in the majority of the cases, a low number of decompositions provides fused images with a high spatial and spectral quality trade-off. Additionally, the results indicate that the decomposition level that provides the best spatial-spectral quality trade-off depends on the spatial frequencies content of the source images.