MuSlP Multi-Sensor Image Processing system G Sawyer*, D C Mason ?, N Hindley*, D G Johnson’, I H Jones-Parry*, C J Oddy$, T K Pike**, T Plassard”‘, A J Rye$, A de Salabert**, B Serpico’+, A L Wielogorski** The MuSIP Multi-Sensor Image Processing project has developed a proof-of-concept software demonstrator for the fusion and analysis of images within a knowledge- based environment. The target applications are the analysis of remotely sensed satellite images for the monitoring of forestry, and the analysis of medical images of the human head. A key element has been the development of data fusion techniques for combining multi-sensor and multi-temporal images. The system architecture is generic and modular, and the system control allows automatic planning and algorithm .scheduling. The system includes a spatial database manager capable of handling very large quantities of raster and vector data within a tiled-image environ- ment, a sophisticated facility for database interrogu- tion, u window-bused user interfuse, and a large set of image processing algorithms. The lutter include algor- ithms for low level processing, image interpretation, automatic image registration, data f&m, and change detection. The system has been implemented on a Sun workstation with selected low level algorithms acceler- uted by u trunsputer array. Ke~w~ords: duta fltsion, automatic plunning, spatial dutubases ‘Marconi Space System Ltd., Anchorage Road, Portsmouth. Hamp- shire PO.? SPU. UK ’ “MBB Space Communications and Propulsion Systems Division. Postfach 80 1 I 69. 80(w) Muenchen 80, Germany iGEC-Marconi Research Ccntre. West Hanningfield Road, Gt. Baddow, Chelmsford CM2 XHN. UK $$Thomson CSF. Laboratoires Elcctroniques de Rennes. Avenue de Belle Fontaine. 35510 Ccsson Sevigne, France +-:Dipartimento di lngegneria Biofisica ed Elettronica (DIBE). Universita di &nova. via Opera Pia l/A, 16145 &nova. Italy $Marconi Command and Control Systems, PO Box 133, Chobham Road. Frimley. Camberley,, Surrey GUI6 SPE. UK $$Hunting Technical Scrwces Ltd., Thamesfield House, Boundary Way. Hemel Hempstead, Hertfordshire HP2 7SR. UK ?Natural Environment Research Council Unit for Thematic Informa- tion Systems (NUTIS), University of Reading. Whiteknights. PO Box 227. Reading RG6 2AB. UK zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA Popr r r c~ ei~ ~ cl: 22 :Mrr~ 1991: zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA revised pcrper rrcrived: 20 Drcernhrr iv91 INTRODUCTION The MuSIP Multi-Sensor Image Processing system is a generic knowledge-based system for 2D multi-source image processing which has been developed within a European ESPRIT II Information Technology project, The two year 29 man-year project has been carried out by an international collaboration of eight partner organizations. A key element of the project has been the development of data fusion techniques for the exploitation of multi-sensor and multi-temporal images and the integration of images with non-image informa- tion. This includes the use of temporal information to improve interpretations of individual data sets and to detect changes in a scene. Demonstrators have been produced illustrating the application of these tech- niques to fused data sets in the two disparate fields of earth resource remote sensing and medical image processing. OBJECTIVES The overall project objective was the development of a general purpose system for automatic image pro- cessing, improved image interpretation and change detection. which was suitable for a wide range of applications, and capable of subsequent system expan- sion. This overall aim encompassed a range of more specific objectives. There were a number of distinct system objectives within MuSIP: 1 Several of these were image processing objectives arising from perceived short-comings in ‘classical’ image processing techniques. l particularly in remote sensing, classical methods still tend to be pixel- rather than region-based. The advantages of a region-based approach for improved scene interpretation have been shown in many studies (see elsewhere’-‘). implying that the system should employ a region-based approach. l classical image processing is still often limited to the analysis of single images. However. for man) 0262-X856/92/009589-21 0 1992 Butterworth-Heinemann Ltd vol 10 no 9 november 1992 5x9