基于KSVM决策树法的人脸检测与定位

Face Detection and Location Based on the Method of Non-linear Support Vector Machines (KSVM) Decision Tree

传统的SVM直接在线性条件下训练SVM分类器完成人脸与非人脸的分类,分类器训练困难,计算量大且速度慢。为构造一个复杂背景下人脸检测与定位的新方法,本文用核函数把SVM推广到非线性SVM,再与二叉树相结合,可以解决多类识别问题,此即KSVM决策树人脸检测方法。在此基础上,人脸面部特征被进一步确认。本文提出了改进的四边界Prewitt边缘算子提取人眼,进而定位五官。实验结果表明该方法检测率较高,虚警率较低,定位准确。

Traditional SVM trains Support Vector Machines classifier directly under linear condition to sort face images and nonface images, and for such kind of classifier, the training is difficult. The amount of compute is big and the speed of training is slow. In order to make a new method of face detection and location in an image with complex background, SVM is extended to non-linear by using kernel functions. Then non-linear Support Vector Machines combine with binomial tree method to solve multi-classification＇s identification ＇problems. It＇s named a method of face detection based on KSVM decision tree. Based on KSVM, face is further identified by face features. And paper presents an improved four-edge Prewitt edge operator to extract eyes in face image, and to locate nose and mouth. Experiments show that, compared with traditional SVM, the accuracy is higher, and the rate of false negative is lower, and location is exacter by using this method.