Face detection has been a research focus of computer vision and pattern recognition for a long time. It also has been applied in many fields, such as automatic face recognition systems, visual surveillance, content-based retrieval, advanced human and computer interaction. However, the complexities of face detection make most algorithms weak in large computation, low efficiency and high rate of false results. Support Vector Machines (SVMs) is a new pattern recognition technology that is established on Statistical Learning Theory. It can solve small-sample learning problems better by using Experiential Risk Minimization. Moreover, this theory can translate the problems of non-linearity space into the linearity space, reducing the algorithm complexity using the kernel function. However, SVMs are still some drawbacks in many practical applications. For example: train speed slow, algorithm complex and decision function operation large, etc. This paper proposed a new method to detect frontal view human faces quickly — the Support Vector Clustering(SVC ) algorithm using the theory of SVM and Statistics. This dissertation adopts background separating technique with skin color detection, combined with getting the person's face area, then training SVC method to detect samples. Since the area is small, this method can guarantee the veracity at the same time acquire the speed near real time. SVC-Based Face Detection is made of training section and detecting section. The training process is that input the chroma information distilling from face samples and non-face samples to train the classifiers, and then get optimal separating hyper plane through adjusting the parameters. And the detecting part is using classifiers to detect faces in images. Emulational examination results prove that this algorithm is reasonable and practical. |