Defining the features-objects and structures-that are of interest in an image, distin-guishing them from the background in the scene, is an important and necessary step before most kinds of measurements (particularly automated measurements) can be performed. There are two quite different approaches used to address this need. One is based on the expectation that there should be some global similarity among the pixels that comprise the features of interest, and some difference between those pixels and others comprising the background. This may be a difference in brightness, color, or texture. That assumption leads to methods that are concerned with finding the optimum thresholds that distinguish the features and separate them from background.