论文部分内容阅读
采用核方法在特征空间推导出一类异于欧氏距离的新度量,代替等度规特征映射(Isomap)中的对噪声敏感的欧式距离,用新度量构造测地距离和相应的最小近邻图,提高Isomap算法的抗噪声能力.利用含噪声的Swiss roll数据和人脸图像数据进行实验验证,结果表明这种基于核特征空间的测地距离具有较强的鲁棒性.
A new metric that is different from the Euclidean distance is deduced in the feature space by using the kernel method instead of the noise sensitive Euclidean distance in the isomap. A new metric is used to construct the geodesic distance and the corresponding minimum neighbor graph To improve the anti-noise ability of Isomap algorithm.Experimental results using Swiss roll data and face image data with noises show that the geodetic distance based on kernel feature space is robust.