2018/10/07 by Shahin Pourbahrami, Pourbahrami, Shahin, Leyli Mohammad Khanli +3
Computer Science · Engineering · #Advanced Image and Video Retrieval Techniques #Medical Imaging and Analysis #Medical Image Segmentation Techniques
paper · pdf · doi:10.48550/arxiv.1810.03084
Due to the increased rate of information in the present era, local\nidentification of similar and related data points by using neighborhood\nconstruction algorithms is highly significant for processing information in\nvarious sciences. Geometric methods are especially useful for their accuracy in\nlocating highly similar neighborhood points using efficient geometric\nstructures. Geometric methods should be examined for each individual point in\nneighborhood data set so that similar groups would be formed. Those algorithms\nare not highly accurate for high dimension of data. Due to the important\nchallenges in data point analysis, we have used geometric method in which the\nApollonius circle is used to achieve high local accuracy with high dimension\ndata. In this paper, we propose a neighborhood construction algorithm, namely\nNeighborhood Construction by Apollonius Region Density (NCARD). In this study,\nthe neighbors of data points are determined using not only the geometric\nstructures, but also the density information. Apollonius circle, one of the\nstate-of-the-art proximity geometry methods, Apollonius circle, is used for\nthis purpose. For efficient clustering, our algorithm works better with high\ndimension of data than the previous methods; it is also able to identify the\nlocal outlier data. We have no prior information about the data in the proposed\nalgorithm. Moreover, after locating similar data points with Apollonius circle,\nwe will extract density and relationship among the points, and a unique and\naccurate neighborhood is created in this way. The proposed algorithm is more\naccurate than the state-of-the-art and well-known algorithms up to almost 8-13%\nin real and artificial data sets.\n