Spatial and Spectral Features
The goal of pattern analysis, in general, is to transform signals into symbolic descriptions [Dud73, Nie90a, Pen86]. For simple classification problems this corresponds to the computation of a class number for an observed signal (c.f. Figure 6.1). Since the amount of data is too high, if images or speech signals are used directly, the signals are transformed into lower dimensional vectors, so called features. Features retain all the information needed for pattern recognition. For image and speech recognition, it is necessary to have a set of features which are, for example, appropriate for the subsequent classification of objects, of single words or for the identification of speakers. The following sections introduce several types of features which can be applied to solve selected classification problems in the fields of image and speech processing.
Unable to display preview. Download preview PDF.