Support Vector Machine And Its Application In Face Detection

Support vector machine is a new machine-learning method developed on statistics learning theory. It shows great advantages in resolving the problems such as small-samples, nonlinear and hyper-high dimension in pattern recognition domain. The real data are often linearly inseparable in the input space. To overcome this, data are mapped into a high dimensional feature space, in which the data are sparse and possibly more separable. In practice, the mapping is also not explicitly given. Instead, a kernel function is incorporated to simplify the computation of the inner product value of the transformed data in the feature space. How to select the appropriate parameters for the kernel function quickly is our research work.Face recognition has many potential applications, such as security systems, human-machine interface, video database search, World Wide Web, etc. Conventional face recognition methods apply one kernel to global features and global features are influenced easily by noise or occlusion, thus the conventional methods are not robust to occlusion.This article mainly contains the following two parts based on the two problems above:(1) For the problem of the selection of the kernel parameter, we use the method which combines the inter-cluster distances in the feature spaces and the mixed kernel function to resolve. Calculating such distance costs much less computation time than training the corresponding SVM classifiers. Thus the proper kernel parameters can be chosen much faster. And the testing accuracy is higher than the standard ones.(2) For the problem of global features are influenced easily by noise or occlusion, we apply the local Gaussian kernels to local features. In this paper, the summation of local Gaussian kernels is used as the integration method. Then we get local Gaussian kernel. After that we use fuzzy support vector machine instead of traditional support vector machine. Experiments shown that, in the case of the partial occlusion, this method has a high recognition rate.