一种网格k-近邻集的边界点识别算法

Boundary Point Recognition Algorithm for Grid k-nearest Neighbor Set

为了高效识别聚类边界,根据边界周围区域存在密度差异的特征,提出了一种网格k-近邻集的边界识别算法(BGN)。在网格空间中,该算法根据网格单元和它最近邻居单元的k-近邻集的质量及其单元间中心距离确定边界度,由边界度和边界阈值判断每个网格单元是否为边界单元或噪声单元。通过从边界单元中提取更靠边缘的数据作为边界点的方式,使得边界更精细。实验结果表明,该算法能有效和快速识别出多密度数据集的聚类边界和噪声。

In order to efficiently identify the cluster boundary,based on the existence of density differences in the surrounding area of the boundary,a boundary point recognition algorithm for Grid k-nearest neighbor set（BGN）is proposed.In the grid space,based on the number of elements of grid cell and its nearest neighbor＇s k-neighbor set,along with the cell-center distance of the unit grids,the boundary degree is determined by this algorithm.According to boundary degree and boundary threshold,this algorithm determines if each unit grid is boundary unit or noise unit.By extracting the data closer to the edge of the boundary to represent as boundary points,this algorithm is capable to make finer boundary.The experimental results indicate that the algorithm can effectively and quickly identify the cluster boundaries and noise for multi-density datasets.