Hyperspectral Data Processing: Algorithm Design and Analysis

32.1 Introduction

Magnetic resonance imaging (MRI) is an advanced medical instrument technology, which provides high contrast of image intensity of information about soft tissues that can be used for image analysis (Sebastiani and Barone, 1991; Wright, 1997). Therefore, one of the fundamental tasks of MRI is tissue classification, which is generally accomplished by segmentation. Technically speaking, classification and segmentation are two completely different concepts. The classification generally requires a set of training samples to perform class membership assignment, which can be carried out in a supervised or an unsupervised manner depending upon how training samples are produceed a priori using prior knowledge or a posteriori obtained directly from the data. On the other hand, segmentation intends to group data samples into a finite number of homogeneous regions, p, according to a certain criterion such as similarity of image intensities, custom-designed features, and so on. Therefore, segmentation is usually performed in an unsupervised manner without assuming any prior knowledge other than the value of p. Many efforts have been devoted to design and development of segmentation algorithms, most of them based on the concept of c-means (k-means) clustering and their various fuzzy versions (Bezdek, 1981; Bezdek et al., 1984). As a result, it is generally referred to as an automatic or unsupervised processing. In order for segmentation to perform classification, a set of training ...