Many applications in image processing rely on robust detection of image features and accurate estimation of their parameters. Features may be too numerous to justify the process of deriving a new detector for each one. This chapter exploits principal components analysis to build a single, flexible, and efficient detection mechanism based on the use of composite rejectors. The complementary aspect of detecting templates considered as a set of separate features will also be addressed presenting an efficient architecture: a rejector cascade classifier built by boosting simple, pixel level classifiers applied to a census transformed image.
keywords: parametric feature manifold, AdaBoost, boosting, census
transform, multi-class pattern rejector, constellation matching