Face detection based on cost-sensitive support vector machines

Abstract

This paper presents a method of detecting faces based on cost-sensitive Support Vector Machines (SVM). In our method, different costs are given to the misclassification of having a face missed and having a false alarm to train the SVM classifiers. The method achieves significant speed-ups over conventional SVM-based methods without reducing detection rate too much and the hierarchical architecture of the detector also reduces the complexity of training of the nonlinear SVM classifier. Experimental results have demonstrated the effectiveness of the method.