Face detection using combinations of classifiers

In this paper we present a two-stage face detection system. The first stage reduces the search space using two heuristics in cascade: 1) in a face image, the average intensity of the eyes is lower than the intensity of the part between the eyes, and 2) the histograms of the grayscale image of a face with uniform lighting have a distinguishable shape. In the second stage we use combinations of different classifiers including: naive Bayes (NB), support vector machine (SVM), voted perceptron (VP), C4.5 rule induction and feedforward artificial neural network (ANN); we also propose a simple lighting correction method. We use the BioID face dataset to test our system achieving up to a 95.13% of correct detections.