Face Detection Based On Multiclassifier Fusion

Human face embodies extremely rich information and is the key symbol for distinguishing, recognizing and memorizing individuals. Human face detection plays an important role in research fields of computer vision, pattern recognition and multimedia technology. As the essential technology in face image application area, human face detection has been researched intensively recently. Supported by National Natural Science Foundation of China, this paper studies the problem of variant pose human faces detection in color image and video.In this paper, the Adaboost cascade face detection algorithm proposed by Viola et al. is analyzed in detail. Firstly, it uses a new image representation called the"Integral Image"which allows the features used by our detector to be computed very quickly. Secondly, a learning algorithm, based on Adaboost, is proposed, which selects a small number of critical visual features from a larger set and yields extremely efficient classifiers. Finally, a method for combining increasing more complex classifiers in a cascade classifier is formed. This algorithm not only decreases false alarm rate and running time significantly, but also improves hit rate and promotes the ability of expansion increasingly. A new face detection algorithm for variant pose faces detection in color image is proposed, which detects faces by using cascade classifiers fusion followed by skin color verification. Two classifiers, one for frontal face and the other for profile face, are trained by the Adaboost algorithm. They work in parallel for color image detection, and their detection results are combined to form candidate face regions. These candidate regions are further verified by a skin color model in YCbCr chrominance space. Finally, the face regions in original image are labeled. The proposed algorithm integrates results from different classifiers, makes use of gray scale and color distribution information of human face and improves hit rate and decreases e false alarm rate significantly. With a large set of color images and videos containing one or more faces with variant pose, the presented algorithm is robust to face pose and background variation.In this paper, three color image test sets are constructed including frontal and profile face test set, variant pose faces test set. Face detection works on these sets using frontal and profile face classifier, combined two types classifier and our proposed classifier respectively. Experiment results show that variant pose face samples are not...