Hyperspectral Data Processing: Algorithm Design and Analysis

12.6 Conclusions

OSP has become a standard hyperspectral imaging technique (Schwengerdt, 1997; Chang, 2003a) that can be used in many versatile applications. Despite the fact that various relationships among OSP, CEM, and the RXD have been studied (Chang, 2003a, 2003b; Du et al., 2003), this chapter investigates many interesting issues resulting from OSP that are not explored in Chang (2003a, 2003b) and Du et al. (2003). It revisits OSP from several signal processing perspectives and offers many insights into its design rationales that have not been investigated previously. In particular, it shows that OSP can be derived from various view points of signal detection, linear discriminant analysis, and parameter estimation where the LS OSP is essentially equivalent to LS-LSMA via the proposed OSP-model. It further studies effects of the Gaussian noise and white noise assumptions on the performance of OSP. Finally, it derives various forms of OSP when OSP is provided by different information levels of target knowledge. As shown in this chapter, some well-known and popular filters such as CEM, TCIMF, and the RX anomaly detector can be considered as members of OSP family. Since many experiments conducted based on real hyperspectral images using various forms of OSP have been reported in the literature and can be also found in Chang (2003a), real hyperspectral image experiments are not included here.