Face Detection Using Adaboosted RVM-based Component Classifier

Abstract

In this paper, a new Adaboosted kernel classifier algorithm is introduced for face detection application. However, most of the methods used to implement relevance vector machine (RVM), need lengthy computation time when faced with a large and complicated dataset. A new pruning method is used to reduce the computational cost. The kernel classifier parameters are adoptively chosen. In addition, using Fisher's criterion, a subset of Haar-like features is selected. As a result, our proposed algorithm with its previous counterparts i.e. support vector machine (SVM) and RVM without boosting is compared, which results in a better performance in terms of generalization, sparsity and real-time behavior for CBCL face database.