ABSTRACT

This chapter is concerned with the task of designing a pattern classifier that assigns class labels to patterns. For example, in the fingerprint recognition problem where patterns correspond to fingerprint images, the labels represent the identity of an individual, and in the disease detection problem where patterns correspond to a patient’s test results, the labels represent disease or not disease. Humans possess a natural ability to recognize spatial and temporal patterns in the world around them by processing signals gathered through their senses. Humans are notoriously bad at certain types of pattern recognition problems where man-made sensors and computers excel. The process of converting input data into patterns relies heavily on the unique characteristics of the application and therefore varies widely from one application to the next. Patterns from individual classes can be highly variable. Indeed, there is often a natural variability within the pattern classes.