Correntropy based scale ICP algorithm for robust point set registration. (September 2019)

• Record Type: Journal Article
• Authors: Wu, Zongze; Chen, Hongchen; Du, Shaoyi; Fu, Minyue; Zhou, Nan; Zheng, Nanning
• Is Part Of: Pattern recognition. Volume 93(2019:Sep.)
• Page Start: 14
• Page End: 24
• Publication Date: 2019-09
• Subjects: Iterative closest point -- Correntropy -- Scale transformation -- Point set registration -- Outliers; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2019.03.013
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: Correntropy is introduced to the scale ICP algorithm, which could eliminate the influence of outliers and noise. The one-to-one correspondence is employed to improve speed. The closed-form solution is given during the transform estimation iteration, which greatly reduces the run-time. It is a general framework for m-dimensional registration, which is independent of shape representation and feature extraction. Abstract: The iterative closest point (ICP) algorithm has the advantage of high accuracy and fast speed for point set registration, but it performs poorly when the point sets have a large number of outliers and noises. To solve this problem, in this paper, a novel robust scale ICP algorithm is proposed by introducing maximum correntropy criterion (MCC) as the similarity measure. As the correntropy has the property of eliminating the interference of outliers and noises compared to the commonly used Euclidean distance, we use it to build a new model for scale registration problem and propose the robust scale ICP algorithm. Similar to the traditional ICP algorithm, this algorithm computes the index mapping of the correspondence and a transformation matrix alternatively, but we restrict the transformation matrix to include only rotation, translation and a scale factor. We show that our algorithm converges monotonously to a local maximum for any given initial parameters. Experiments on synthetic and real datasets demonstrate that the proposed algorithm greatly outperforms state-of-the-art methods in terms of matching accuracy and run-time, especially when the data contain severe outliers.