Two-Stage Outlier Elimination for Robust Curve and Surface Fitting. (29th June 2010)

• Record Type: Journal Article
• Authors: Yu Yu, Jieqi Jieqi; Zheng Zheng, Haipeng Haipeng; Kulkarni Kulkarni, Sanjeev R. Sanjeev R.; Poor Poor, H. Vincent H. Vincent
• Other Names: Shao Shao Ling Ling Academic Editor.
• Is Part Of: EURASIP journal on advances in signal processing. Volume 2010(2010)
• Publication Date: 2010-06-29
• Subjects: Signal processing -- Periodicals; Traitement du signal; Signal processing; Periodicals; 621.3822
• DOI: 10.1155/2010/154891
• Languages: English
• ISSNs: 1687-6172

Abstract

Abstract : An outlier elimination algorithm for curve/surface fitting is proposed. This two-stage hybrid algorithm employs a proximity-based outlier detection algorithm, followed by a model-based one. First, a proximity graph is generated. Depending on the use of a hard/soft threshold of the connectivity of observations, two algorithms are developed, one graph-component-based and the other eigenspace-based. Second, a model-based algorithm, taking the classification of inliers/outliers of the first stage as its initial state, iteratively refits and retests the observations with respect to the curve/surface model until convergence. These two stages compensate for each other so that outliers of various types can be eliminated with a reasonable amount of computation. Compared to other algorithms, this hybrid algorithm considerably improves the robustness of ellipse/ellipsoid fitting for scenarios with large portions of outliers and high levels of inlier noise, as demonstrated by extensive simulations.