Unsupervised and Supervised Segmentation Methods

Other classical unsupervised cluster analysis methods rely on using mathematical indicators, such as distances, to quantify the similarity among pixel spectra. Thus, each pixel can be viewed as a point in the space of original wavenumbers or on other spaces, for example PC space. The coordinates of a pixel can be the spectral readings at the different wavenumbers (in the original image space) or the scores (in the PC space). Similar pixels should be close in the reference space and, therefore, distance measurements, such as Euclidean distance ( ), can be used to assess this proximity  

Although supervised and unsupervised methods have been in the past well-separated families, a number of hybrid approaches have appeared that combine both unsupervised and supervised classifiers for image segmentation purposes [61]. 

Resuits obtained using (a) raw pixel spectra, (b) PCA scores, and (c) MCR scores. (Reprinted from Ref. [19]. Eisevier.) 

Supervised segmentation methods work defining the different pixel classes beforehand with a series of well-identified pixels. Each different class is described by a model and these models are used to assign unknown pixels to the predefined classes. Reference pixels in and out of the classes can be selected by applying class membership masks in the image score scatter plots [74, 75], for instance, or by using results from unsupervised segmentation methods. 

Supervised methods are very often integrated in the software of hyperspectral acquisition systems, and once the classification model is built for a particular problem, this model acts automatically in new acquired images and allows for in situ analysis (e.g., in portable imaging devices to distinguish acceptable fruits or not [84]) or for online control for pushbroom imaging systems (e.g., in sorting production lines for food or other materials [60, 85]). 

---