Features-Based K-views Model

This chapter describes a new K-views algorithm, the K-views rotation-invariant features (K-views-R) algorithm, for texture image classification using rotation-invariant features. These features are statistically derived from a set of characteristic views for each texture. Unlike the basic K-views model such as K-views-T method, all the views used are transformed into rotation-invariant features, and the characteristic views (i.e., K-views) are selected randomly. This is in contrast to the basic K-views model that uses the K-means algorithm for choosing a set of characteristic views (i.e., K-views). In this new algorithm, the decision of assigning a pixel to a texture class is made by considering all those views, which have the pixel (being classified) located inside the boundary of their views. To preserve the primitive information of a texture class as much as possible, the new algorithm randomly selects K-views of the view set from each sample sub-image as the set of characteristic views.