Multidimensional fusion by image mosaics 213 era vignetting effects that were calibrated beforehand. We registered the images using a method discussed in [32]. The registration yielded a wide multispectral image mosaic, where the spectrum can be computed for each point. The multispectral mosaic was then converted to a colour mosaic, shown in Figure 8.18. Similarly to Sections 8.2 and 8.3, a full range spectrum is obtained in the central region of interest. This region seems yel- lowish in Figure 8.18 because the light coming from the incandescent lamps is rather yellow. 5 Other than that, the estimated colours were consistent with the appearance of the objects. Information is obtained also about the periphery, though with decreasing spectral range. The top of Figure 8.18 indeed shows the periphery regions, one frame wide, on both sides of the mosaic central part. Since the spectral range in these regions changes gradually from the central part, there is no abrupt decrease of quality in the periphery. However, towards the outer boundaries of the left and right periphery, the image becomes red and blue respectively. This is due to the absence of data on the complimentary wavelengths in these regions [32]. Even there, substantial information can still be available for algorithms that make do with partial spectral data, or that do not rely on colour but on spatial features. For example, the objects in the right periphery clearly appear in the raw frame shown at the last photograph in Figure 8.17. It shows [32] loose dark cables hanging down through the frame, and their shadows on the wall behind. This can indicate the number and spatial distribution of the light sources in the scene, as in [74]. In addition, other objects (shaped bricks) can be recognised in this region. Therefore, the peripheral regions are not wasted