Department of Mathematical Modelling, Technical University of Denmark
Building 321, DK-2800 Lyngby, Denmark
Phone +45 4588 1433, Direct +45 4525 3413, Fax +45 4588 1397
E-mail email@example.com, Internet www.imm.dtu.dk
In remote sensing there is usually a trade-off between spatial resolution and spectral resolution. It is often desirable to present a combination of the data which merges the low resolution spectral (colour) information with the high resolution spatial information. If the ratio in spatial resolution between the greyscale image and the colour image becomes too great the image resulting from usual itensity substitution techniques will tend to look as if it has "squares" of colour information overlaid on the greyscale image.
A new way of "fusing" the colour information with the greyscale information is presented here. The technique is based on Markov random field (MRF) assumptions and the maximum a posteriori (MAP) estimate is found by means of iterated conditional modes (ICM). The design of the specific Markov random field used here allows the user to control three properties of the resulting "merged" image: 1) degree of colour smoothness between neighbouring pixels, 2) degree of confidence in the original colour information (e.g. SPOT HRV XS), and 3) degree of confidence in the original panchromatic information (e.g. orthophoto). As a possible fourth control element, we may exclude neighbours belonging to other segments of the segmented orthophoto from the neighbourhood of the MRF.
The data used here as an example are from the town of Sorø in Denmark. The data consist of a digital orthophoto (62.5 cm pixels) and a SPOT HRV XS (20 m pixels) giving a linear ratio of 32:1 between orthophoto and SPOT pixels. In this presentation, we compare results based on neighbourhoods with and without the exclusion of neighbours in other segments.