关键词: Adaboost, BitBP特征, 多重级联, 人脸检测

Abstract:

This paper proposes a distinctive feature extraction method using Bit Binary Pattern (hereafter referred to as BitBP) to represent facial image for frontal face detection.The BitBP feature encodes the integral image of rectangular regions by binary pattern operator.An effective face classifier is trained with the RealAdaboost algorithm and LUT weak classifier for front face detection.Experiments results demonstrate that the new BitBP feature is more distinctive than Haar-like feature and MB-LBP feature.Given a false positive rate of 0.1%,the classifier based on BitBP features shows 2% higher correct detection rate than the Haar-like feature and 0.14% higher correct detection rate than the MB-LBP feature.The training time is also efficiently reduced since the BitBP feature has a much smaller size.A cascade of classifiers which are already cascaded is constructed into the final classifier,and all the classifiers are based on BitBP features.

Key words: Adaboost;BitBP feature;multi-cascade;face detection