A PCA-fuzzy clustering algorithm for contours analysis

Abstract

Principal component analysis (PCA) is a usefully tool for data compression and information extraction. It is often utilized in point cloud processing as it provides an efficient method to approximate local point properties through the examination of the local neighborhoods. This process does sometimes suffer from the assumption that the neighborhood contains only a single surface, when it may contain curved surface or multiple discrete surface entities, as well as relating the properties from PCA to real world attributes. This paper will present a new method that joins the fuzzy clustering algorithm with a local sliding PCA analysis to identify the non-linear relations and to obtain morphological information of the data. The proposed PCA-Fuzzy algorithm is performed on the neighborhood of the cluster center and normal approximations in order to estimate a tangent surface and the radius of the curvature that characterizes the trend and curvature of the data points or contour regions. © 2011 Springer-Verlag Berlin Heidelberg.