Principal curves algorithm usually performed not well in initializing, so it ended up with a low accuracy of topology structure. In order to find the primary topology of data, the improved algorithm replaced principal component line with connect K-nearest neighbors in initializing. Tested the improved algorithm by applying to skeletonization of off-line handwritten digits.The results show that the improved algorithm overcomes the shortcoming above-mentioned. It turns out that the improved algorithm performs better than existing methods in computational complexity, connective, parameter, and it is more capable in handling connective data.%针对主曲线算法初始化步骤效果差、导致所得结果不能正确反映数据的拓扑结构问题,对其进行改进,用连通K近邻代替第一主成分线找到数据的初始拓扑结构.将改进后的算法应用于脱机手写体数字骨架化,实验证明,改进的算法克服了上述缺点,能更好地找到数据的拓扑结构,在时间复杂度、连通性、参数等方面均优于其他改进算法.改进算法适于具有"连通性"的数据.
展开▼