ABSTRACT
In many applications, it is necessary and useful to represent data in a standard format. This enables comparing the feature values of an object from one dataset directly to another object from another dataset. A common example is images where two feature vectors produced from two different images may need to be compared. The dimensionality reduction techniques such as PCA (Section 9.5) and SVD (Section 9.4) project on basis vectors that are produced from the datasets themselves. Thus, two datasets will have different projection vectors. The data representation techniques described in this chapter use a consistent set of basis vectors for projection. As a side effect, these methods can be used as dimensionality reduction tools as well. We first describe three discrete signal processing methods before concluding with a histogram representation technique. The signal processing methods work well for time-series data.
