IGARSS’06 Data Fusion Contest: Fusion of Panchromatic and Multispectral Images. Luciano Alparone 1 , Lucien Wald 2 , Jocelyn Chanussot 3 , Lori Man Bruce 4 and Paolo Gamba 5 1 University of Florence, Italy (alparone@lci.det.unifi.it) 2 Ecole des Mines de Paris, France (lucien.wald@ensmp.fr) 3 Grenoble National Polytechnic Institute, France (jocelyn.chanussot@lis.inpg.fr) 4 Mississippi State University, USA (bruce@ece.msstate.edu) 5 University of Pavia, Italy (paolo.gamba@unipv.it) Abstract— The IEEE Data Fusion Technical Commit- tee is organizing a “Data Fusion Contest” in conjunction with the IGARS Symposium. For the first issue of the contest, the fusion of multispectral and panchromatic images for optical very high resolution images is consid- ered. A comparative study of the results provided by the participants on common data sets is presented during the conference in the “Data Fusion” session and the “winning” team(s) is (are) granted an IEEE award during the IEEE GRS-S Technical Committee and Chapter Dinner. I. I NTRODUCTION Quoting [1], data fusion can be defined in a general way as a formal framework in which are expressed the means and tools for the alliance of data originating from different sources. In particular, the fusion of images having different spatial resolutions is a very important issue. The interest for such techniques has been recently reinvigorated by the development of new optical sensors providing panchromatic data with a very fine resolution (typically one meter or less) together with some multispectral images with a lower spatial resolution (typically a few meters). To stimulate research activity in this field and provide a comparative state of the art, the IEEE Data Fusion Technical Committee is organizing a “Data Fusion Contest” in conjunction with the IGARS Symposium. For this first issue of the contest, the fusion of multispectral and panchromatic images for optical very high resolution images is considered. In the recent literature, numerous methods have been proposed: • multiresolution approaches, from the classical wavelet-based strategies, to more recent curvelet- [2] [3] or ridgelet-based algorithms [4]. These ap- proaches are usually considered as well preserving the spatial features while sometimes introducing spectral distortion. • other strategies involve the intensity-hue- saturation (IHS) representation of color and usually ensure spectral consitency [5] [6]. • some hybrid methods using the IHS representa- tion, together with PCA mergers based on wavelet decomposition have also been proposed [7] Some interesting comparative studies have already been published [8] [9] [10] [11]. For this contest, all the participants will test their best algorithms on common data sets. Beyond a subjective analysis, the results will be objectively evaluated using the metrics proposed by Alparone in [12]. This paper will present the results and a comparative analysis of the pansharpened images provided by all the participants. II. DATA SETS Various data sets have been provided to the partici- pants: • six Pleiades simulated images (a future satellite to be launched in 2009) have been provided by CNES, the French National Space Agency. Each data set includes a very high panchromatic image (80 cm resolution) with corresponding multi-spectral images (3.2 m resolution). An air- borne multispectral very high resolution image was available as a ground truth. It has not been distributed to the participants but was used by the organizing committee for the evaluation of the results. • some QUICKBIRD images have also been pro- vided by the Mississippi State University. The data were provided with the original resolution, and a downsampled version. The participants were requested to process both data sets independently. III. SCHEDULE The contest started as early as december 2005 with the constitution of the organizing committee, including the IEEE GRS-S Data Fusion Technical Committee chair and co-chair, as well as Guest Organizers. After the definition of the aim and “rules” of the contest and collection of the data, the contest has been advertized