The Study Of Boundary Detecting Algorithm Based On The Two-stage Technique

With the development and innovation of science and technology, information and technology and computer technology has brought our society great progress. The amount of information of the database is going along with the rapid growth, and they are forming into a large number of heterogeneous data which contains much informiation. Data mining is a fast and effective technology that can help us to extract useful information from the multiple data. Cluster analysis is an important part of data mining. It is very common in any analysis with multivariate data field. Boundary points detecting of clusters plays an important role in image processing, machine learning and the practical application. However, the cluster boundary detection research is still in its infancy. Accurately detecting the cluster boundary will help to improve the accuracy of clustering and classification. It also can help to study deeply the characteristics of the boundary data.In this paper, based on the analysis of the existing cluster boundary detection algorithm, we have the idea of stepwise refinement. First, we obtain candidate boundary set, and then we determine the precise boundary of clustering. So two new clustering boundary detection algorithm have been proposed:BDDTS (Boundary Detection algorithm of clusters based on Dual Threshold Segmentation) and BDHO(Boundary Detection algorithm of clusters based on improved Harris Operator).The algorithm BDDTS, accruing to the different cost function values of data points, divides data set into internal point set, intermediate point set and external point set. After removing the internal points from intermediate point and combining it to external point set, we get candidate boundary set. At last, we do the secondary treatment to the candidate boundary set and then obtain more accurate boundary. The experimental results show that BDDTS can detect boundary points of clusters in arbitrary shapes, sizes, density very rapidly and efficiently. It is also applicable to real data set and high-dimensional data set. At the same time, the algorithm parameters are easily selected.The algorithm BDHO is basing on harris operator in image processing. First we select a reference direction. And then we select another direction through the angle information. Calculate the Direction Change Rate of the two directions. According to the improved harris operator, we calculate the Boundary-Response Function value. And use the boundary-response threshold to remove most of the internal points and noises. The left data points are part points near the boundary and sparse part of the internal point. At last we make use of angle-response threshold to improve this boundary. The algorithm can handle arbitrary shapes, sizes, density data bases.The two boundary detection algorithms proposed in this paper have been introduced from other areas of knowledge. Both of them can effectively solution the problem of boundary detection in cluster analysis.