Dataset background

Dataset background

For the design and evaluation of artificial pattern recognition systems sets of objects given by some data representation are needed. PRTools is based on the vector representation of objects. Traditionally this is realized by features: object properties that are useful for distinguishing and recognizing classes in sets of objects. PRTools offers various other ways to represent objects by vectors: pixels or samples (of images or time signals), dissimilarities, kernels, class posterior probabilities. They are of primary significance in designing and using pattern recognition systems.

To facilitate the use of sets of feature vectors in the context of pattern recognition problems PRTools offers a Matlab object oriented programming class called dataset ¹. A variable of the type dataset stores internally all feature vectors of a set of objects. In addition various kinds of annotation can be stored in such a variable as well: object labels, class names, feature names, prior probabilities, etcetera.

Dataset background