Compression of Spectral Images

In this chapter we describe methods how to compress spectral imaging data. Normally the spectral data is presented as spectral images which can be considered as generalizations of colour images. Rapid technological development in spectral imaging devices has initiated the need for the compression of raw data. Spectral imaging has been central to many remote sensing applications like geology and environment monitoring. Nowadays, new application areas have arisen in industry, for example in the quality control of assembly line products and in applications, where the traditional three-chromaticity colour measurements are not accurate enough. Spectral imaging produces large amounts of raw data which will be processed later in various applications. Image compression provides a possibility to reduce the amount of raw data for storing and transmission purposes. The image compression can be either lossless or lossy. In the lossy compression the quality of the reconstructed data should be estimated to evaluate the usefullness of the reconstructed data. The lossy compression is justified in the sense that the compression ratios are much higher than in the lossless case where the reconstructed data is identical to the raw data. Spectral images are now available for different applications due to the development in the spectral imaging systems (Hauta-Kasari et al., 1999; Hyvärinen et al., 1998). Geoscience and remote sensing have been the main application areas of spectral images but nowadays several new application areas have emerged in industry. For example, applications in quality control, exact colour measurement, and colour reproduction use spectral information, since RGB colour information only is not sufficient. Image compression has been one of the main research topics in image processing. The compression methods are usually developed for images visible to humans, i.e. for grey-scale or RGB colour images. Applications in the field of remote sensing and recent advances in industrial applications, however require the compression of spectral images (Vaughn & Wilkinson, 1995). Some compression methods are lossless (Memon et al., 1994; Roger & Cavenor, 1996), but most of the methods are lossy (Abousleman et al., 1997; Gelli & Poggi, 1999). Some applications can accept data which is compressed by a lossy scheme, but naturally the important features in the data must be present. If the lossy compression method cancels out any of the important features for the applications, then the lossless compression is the only possibility to decrease the amount of the raw data. Compression is required due to the large amounts of data captured in the images. Regular digital cameras in everyday use apply JPEG or TIFF-compression. Images displayed in web-