Joint boosting of histogram like features for the generic recognition of object classes and subclasses

A very important aspect of visual human computer interaction is the robust and effective recognition and localization of multiple object classes in the visual input data. This paper addresses the problem of generic object class and sub-class classification. Vector valued histogram-based image features are used in a joint boosting classifier to provide an efficient multi-class object detector. A novel weak learner based on Multiple Discriminant Analysis is introduced for vector valued histogram features which allows to combine them in a multi-class boosting based classifier. Successful experimental results on a publicly available dataset proves the feasibility of the proposed approach.