Feature Generation II

This chapter examines feature generation with a focus on image and audio classification. Feature generation is a procedure that computes new variables that in one way or another originate from the stored values of the image array I ( m , n ). Some of the feature generation techniques can be considered common and can be applicable in both visual and audio modalities. A large number of features are the result of different approaches to exploit the specific nature of the signals and encode the required classification information in a more efficient way. This chapter also focuses on first- and second-order statistics features as well as the run-length method. The chapter discusses typical features used to characterize and classify audio information. The chapter presents a description of statistical properties of signals and images and the ways these can be exploited to extract information-rich features for classification. The chapter examines whether the notion of self-similarity is extendable to stochastic processes and, if it is, how useful it can be. The chain code for shape description is discussed in this chapter. Computer exercises are then offered to generate these features and use them for classification for some case studies.