Abstract Point set registration is a fundamental technique in machine learning and data mining, crucial for tasks such as computer vision, pattern recognition, and robotics. Traditional methods, including the iterative closest point (ICP) algorithm, often struggle with noisy data and outliers. This paper introduces a novel rigid registration algorithm based on the maximum correntropy criterion (MCC), which integrates correntropy into the point-to-plane (ptDpl) distance measure, proposing a new energy function. An iterative process is employed to minimize this energy function, computing correspondences and transformation parameters at each step. The algorithm converges monotonically to a predetermined local maximum, demonstrating robustness against cutting, noise, and outliers. Experimental results on synthetic and real-world data show that our 2D- and 3D-based method outperforms the traditional ICP algorithm and the Welsch criterion in accuracy and efficiency. Relevant codes and data are published to https://github.com/ygq0000/mcc
Yang et al. (Mon,) studied this question.