This chapter presents two major hyperspectral imaging techniques, orthogonal subspace projection (OSP) and constrained energy minimization (CEM), which have recently found success in spectral signature detection in magnetic resonance (MR) images. Unlike classical image classification techniques which are primarily designed on a pure pixel basis, the proposed OSP and CEM are mixed pixel classification techniques which estimate abundance fractions of different material substances assumed to be present in the image data so as to achieve classification. Three scenarios are considered for MR image classification based on different levels of object information provided a priori. The OSP is applied to the case that the complete prior object knowledge is known compared to the CEM which is used for the case that only desired object knowledge is given while other knowledge can be discarded. When no prior object knowledge is available, the OSP is extended to an unsupervised OSP (UOSP), which obtains the necessary object knowledge directly from the image data in an unsupervised manner for OSP classification. In order to demonstrate the utility of these three scenarios in MR image classification, a series of experiments are conducted and compared to the commonly used c-means method for performance evaluation. The results show that both OSP and CEM are promising and effective spectral techniques for MR image classification.
All Science Journal Classification (ASJC) codes