Color in Computer Vision: Fundamentals and Applications

Chapter 18: Segmentation of Multispectral Images

With contributions by Harro M. G. Stokman

Spectral information has become an important quality factor in many imaging processes because of its high accuracy. ¹Spectral imaging is used, for example, in remote sensing, computer vision, and industrial applications. Spectral images can be obtained, for example, by a CCD camera with narrow-band interference filters [4]. Photometric invariance can be derived from multispectral images. In fact, the techniques presented in the previous chapters can be used to detect regions in multispectral images. To obtain robustness against noise, noise propagation can be adopted as discussed in Chapter 4. More information can be found in Reference 5.

In this chapter, methods are discussed to obtain photometric invariant region detection. In Section 18.3, the effect of sensor noise is discussed. Region detection is described in Section 18.4. In Section 18.5, the theoretical estimated uncertainty in polar angular representation is compared empirically to the real uncertainty. Experiments are carried out to evaluate the segmentation method, which are discussed in Section 18.6.

18.1 Reflection and Camera Models

In this section, we discuss the camera and image formation model. On the basis of the models, we examine cluster shapes drawn by uniformly colored objects in multispectral color space.

18.1.1 Multispectral Imaging

We use the Imspector V7 spectrograph from Spectral Imaging Ltd. The spectrograph transforms ...