A Probabilistic Approach to Adjust Adaboost Weights

Abstract

This work presents a probabilistic modified Adaboost algorithm for face detection. The goal of the proposed approach is to establish a detection algorithm that reduces false positive detection rate. As our contribution we build a new weighting system which considers both weak classifier total error and classification probability. The probability is determined by computing both positive and negative images classification probabilities by each weak classifier. The new weighting system gives higher weights to weak classifiers with best positive images classifications and this reduces false positives. Experiment results have revealed that the original Adaboost and the proposed method have comparable face detection rate performances while false positives are almost four times reduces using the proposed method.