Multispectral imaging is applied in the context of various applications. Recent technological advances, like the development of low-cost and compact multispectral imaging cameras, show the ongoing popularity of this imaging technique. Because of this popularity, there is a high demand for generic solutions to enable a user to get access to relevant information, e.g. constituent spectra, for a large variety of application domains. But, due to the typically high-density of spectral information, usually the interpretation, on the one hand, is complex for humans and on the other hand, is time-consuming for computers. Thus, the challenging task of interpretation consequently requires efficient data analysis algorithms and intuitive visualization methods to support the understanding of the data and to finally make use out of them.
This thesis addresses the identified challenges mainly by the presentation of efficient, intuitive and generic visual analysis methods for both, multispectral image segmentation and linear spectral unmixing. The key function of these methods is the involvement of a user by visual feedback to enhance the guidance in the exploration. Based on the visual support, the user is enabled to explore and assess results of automatic analysis algorithms and to optimize them, if necessary.
In addition to the visual analysis methods and with the aim to provide a further understanding of the processing of multispectral image data, also more fundamental challenges and concepts of the processing are discussed.