基于尺度不变特征变换和SVM的人脸检测算法

A FACE DETECTION ALGORITHM BASED ON SCALE-INVARIANT FEATURE TRANSFORM AND SUPPORT VECTOR MACHINE

模糊支持向量机的提出克服了过学习问题和减少了多类问题分类时存在的不可分区域,被成功地应用在分类检测问题中。在人脸的特征提取和检测实验中,使用变分的测地活动轮廓模型对人脸分割定位,通过尺度不变特征变换算法提取人脸图像的数字特征,然后用一种改进的紧密度模糊支持向量机进行人脸检测。改进的紧密度模糊支持向量机通过交叉确认来获取较好的隶属度函数的参数,提高了算法的执行效果。实验表明改进算法有较好的分类精度和鲁棒性。

The presentation of fuzzy support vector machine overcomes the overfitting problem and reduces the unclassifiable regions existing in multi-class classification, and has been successfully applied in classification and detection issues. In experiments of face features extraction and detection, the variational g6odesic active contour model is used for face segmentation and localisation, and scale-invariant feature transform algorithm is employed to extract the digital features of face images. Then an improved affinity fuzzy support vector machine is applied to face detection. This improved affinity fuzzy support vector machine obtains better parameters of membership functions through cross-validation, which improves the implementation effect of the algorithm. Experiments show that the improved algorithm has better classification accuracy and robustness.