Research And Implementation Of Asymmetric Face Detection Algorithm

In recent years, because of the development of advanced technology, face recognitionengages lots of attention and interest. It is widely used in the field of security and becomesan active research issue. Face detection is both a core factor in face recognition and atypical case in target detection. Its performance directly affects the overall performance ofthe face recognition system. Currently, face detection has also gotten the widespreadattention.This thesis studies Viola and Jones’s AdaBoost algorithm and goes into details for itstheory. LAC algorithm is used to solve the asymmetric problem in face detection. Thisthesis achieves an asymmetric face detection algorithm. First of all, fast trainingAdaBoost selects rectangular features and computes the threshold. Then LAC algorithmsets the learning goal to build a strong classifier that is combined by the selected features.Finally, the face detection classifier is cascaded by those strong classifiers. This methodcan availably decrease the cost of the weak classifiers and solves the asymmetric problem.The training process of the cascade classifier and the management of the training samplesare introduced in detail. And eyes detection is added to make sure to reduce the falsepositive rate of face detection.The results show that the asymmetric face detection algorithm of fast trainingbalances the training time and the performance. Not only can the classifier get highdetection rate, but also get shorter training time and solve the asymmetric problem. It cansatisfy the practical application for face detection with eyes detection.