Various Unsatisfactory Definitions Of Information Fusion

Various definitions of data fusion are proposed in the literature. This item reviews several of them and highlights their advantages, drawbacks and shortcomings. The definition of the JDL is the subject of a separate document.

A few definitions of data fusion or information fusion can be found in the literature, apart that of the JDL discussed in another document. In geography, including images from airborne or spaceborne instruments and analysis of collected intelligence, the documents of the Open GIS consortium define fusion as "the process of organizing, merging and linking disparate information elements (e.g., map features, images, text reports, video, etc.) to produce a consistent and understandable representation of an actual or hypothetical set of objects and/or events in space and time". In these documents, fusion is clearly a set of algorithms, techniques and operators. Fusion is conceived mostly as an analyst-driven process. They further define merging and integration as the process of physically merging two data sets into a common, or fused, representation".

In Earth observation from space, Pohl, Van Genderen (1994) proposed "image fusion is the combination of two or more different images to form a new image by using a certain algorithm", which is restricted to images. Mangolini (1994) extended data fusion to information in general and added a reference to quality. He defined data fusion as a "set of methods, tools and means using data coming from various sources of different nature, in order to increase the quality (in a broad sense) of the requested information". These definitions put the accent on the methods. They contain the large diversity of tools, but are restricted to these.

In applied mathematics and image processing, the definition proposed by Hall, Llinas (1997) also refers to information quality and details the purposes of the data fusion. But it still focus on the methods: "data fusion techniques combine data from multiple sensors, and related information from associated databases, to achieve improved accuracy and more specific inferences that could be achieved by the use of a single sensor alone". Li et al. (1993) wrote " fusion refers to the combination of a group of sensors with the objective of producing a single signal of greater quality and reliability". Quality and reliability are referred to, but there is no reference to concepts. Furthermore it is restricted to sensors and signal.

Data fusion itself is not very recent. It has been created to better answer concerns of Defence agencies approximately twenty years ago, and it is slowly spreading to other domains such as remote sensing and environment. By no means, it brought a breakthrought in these domains up to now. In remote sensing, data fusion processes, such as classification techniques for mapping, are performed since long, without naming it, and even less looking at it as a concept. According to discussions, remote sensing specialists often say that data fusion is fully illustrated by the merging (the word "combination"can also be used) of images having several spectral bands but a low spatial resolution with a panchromatic images having a higher spatial resolution. Examples are SPOT-XS and P, and Landsat 7-ETM and P. The IHS method is one of the most used methods to perform such an operation (Pohl, Van Genderen 1998). It may also be used to combine radar and optical imagery.

This common view of data fusion certainly helps in maintaining a narrow perception of what data fusion is. It also helps in maintaining a three-levels description of data fusion: pixel-level, feature-level and decision-level, wich is completely artificial and in fact opposite to the concept of data fusion (Pau 1988; Wald 1999). Indeed, it is often written that fusion takes place at three levels in fusion of images and data fusion: pixel, attribute and decision. It presents two drawbacks. The word "pixel" is inappropriate here ; the pixel is only a support of information and has no semantic significance. Measurements or observations or signal would be more appropriate. But overall, such a categorization may be misleading: it may falsely imply that fusion processes do not deal simultaneously with these different levels. In Earth observation domain, one may use some features (attribute level) held in a geographical information system to help in classifying multispectral images (measurement level) provided by several sensors. In this particular case, some data are measurements of energy, and others may be symbols. The formalism of Houzelle, Giraudon (1994) allows all semantic levels (measurements, attributes, decisions) to be simultaneous inputs of a fusion process. Wald (1998) presented several examples of this formalism applied to remote sensing.

Indeed most of these definitions are focusing too many on methods though paying some attention to quality. As a whole, there is no reference to concept in these definitions while the need for a conceptual framework was clearly expressed by the scientists as well as practitioners.

References:

D. L. Hall, and J. Llinas. An introduction to multisensor data fusion. In Proceedings of the IEEE, vol. 85, n° 1, pp. 6-23, 1997.

S. Houzelle, and G. Giraudon. Contribution to multisensor fusion formalization. Robotics and Autonomous Systems, vol. 13, pp. 69-85, 1994.

H. Li, B. S. Manjunath, and S. K. Mitra. Multisensor image fusion using the wavelet transform. Computer Vision, Graphics, and Image Processing: Graphical Models and Image Processing, vol. 57, pp. 235-245, 1993.

M. Mangolini. Apport de la fusion d'images satellitaires multicapteurs au niveau pixel en télédétection et photo-interprétation. Thèse de Doctorat, Université Nice - Sophia Antipolis, France, 174 p., 1994.

OpenGIS, Geospatial fusion services testbed. The Open GIS Consortium (OGC), Wayland, Ma, USA, 2000.

L. F. Pau. Sensor data fusion. Journal of Intelligent and Robotics Systems, vol. 1, pp. 103-116, 1988.

C. Pohl, and J. L. van Genderen. Multisensor image fusion in remote sensing: concepts, methods and applications. International Journal of Remote Sensing, vol. 19, n° 5, pp. 823-854, 1998.

L. Wald. An overview of concepts in fusion of Earth data. In Proceedings, EARSeL Symposium 1997 "Future Trends in Remote Sensing", Lyngby, Denmark, P. Gudmansen Ed., A. A. Balkema Publ., Rotterdam, pp. 385-390, 1998.

Wald L., Some terms of reference in data fusion. IEEE Transactions on Geosciences and Remote Sensing, 37, 3, 1190-1193, 1999.

31-08-2001 - Copyright L. Wald, Armines / Ecole des Mines de Paris