Change detection from pan-sharpened images: a comparative analysis

Multitemporal analysis of very high resolution images has gained an ever increasing attention due to the availability of several satellite platforms with different spectral and geometrical resolutions and revisit times. The last generation of multispectral sensors (e.g., Quickbird, Ikonos, SPOT-5) can acquire a panchromatic (Pan) image characterized by a very high geometrical resolution and a set of multispectral (MS) images that have lower spatial resolution. In order to merge the properties of these two kinds of data, i.e. to achieve a set of MS images with an enhanced geometrical resolution, it is possible to use proper multiresolution fusion (merging) techniques, usually called pan-sharpening. The goal of this paper is to analyze the effects of multiresolution fusion on multitemporal MS and Pan images for applications of change detection (CD). The rationale of the analysis consists in understanding in what conditions and to which extent the merging process can improve the results of a standard unsupervised method of change detection. In order to properly analyze and study this problem, different multiresolution fusion algorithms are tested to compare the CD results obtained from original MS and Pan data and from spatially-enhanced MS data. To this aim, both consolidated widely-used algorithms and more recent advanced methods are considered. The following techniques are selected and compared: IHS-like methods [1], the algorithm based on the à-trous wavelet proposed in [2], and the pan-sharpening method based on a minimum mean squared error approach described in [3]. In this way, different methodological approaches to pan-sharpening are taken into account: component-substitution methods, multiresolution analysis (MRA) methods, and hybrid methods. From a theoretical point of view, the algorithms strictly based on standard component substitution locally introduce spectral/radiometric distortions during the fusion process which may negatively affect the performances of unsupervised CD methods. On the other hand, MRA-based methods performing a highpass detail injection may produce spatial distortions, typically ringing or aliasing effects, originating shifts or blur of contours and textures. These drawbacks, which may be as much annoying as spectral distortions for CD applications, are emphasized by misregistration between MS and Pan data, especially (but not exclusively) if the MRA underlying detail injection is not shift-invariant. Efficient hybrid methods should represent a good trade-off between component-substitution and MRA-based pan-sharpening approaches. In order to properly understand the impact of pan-sharpening on the change-detection process, the CD step is performed according to a standard change vector analysis (CVA) technique [4],[5]. The CVA …