Multidimensional Features Extraction Methods in Frequency Domain

Pattern recognition have been a topic of active research during the 30 years, due to the high performance that these schemes presents, when they have been used in the solution of many practical problems in several fields of science, medicine and engineering. The efficiency of pattern recognition algorithms strongly depends in an accurate features extraction scheme that be able to represent the pattern under analysis using a number of parameters as small as possible, while keeping a large intra-pattern and very low inter-pattern similarities. These requirements have led to the development of several feature extraction methods, which can be divided in three groups. Feature extraction methods in time domain, spatial domain and frequency domain. In all cases the proposed feature extraction methods strongly depend of the specify applications. Thus the features extraction methods performing well in some applications, may do not perform well in others, for example the features extraction methods used for speech or speaker recognition are quite different to those used for fingerprints or face recognition. This chapter presents an analysis of some successful frequency domain feature extraction methods that have been proposed for applications involving audio, speech and images pattern recognition. Evaluation results are also provided to show the effectiveness of such feature extraction methods.