Bidirectional Maximum Correntropy-based Isotropic Scaling Registration Algorithm

Point set registration is a core problem in 3D vision and robotics. However, in complex scenarios such as low overlap, noise interference, and scale variations, traditional ICP algorithms are susceptible to outliers, leading to registration failure. To address this issue, this paper proposes a Bidirectional Maximum Correntropy Criterion Scaling Iterative Closest Point (Bi-MCCSICP) algorithm, which integrates bidirectional nearest neighbor search with the maximum correntropy criterion (MCC) to achieve highly robust and accurate point set registration. First, the algorithm introduces a bidirectional nearest neighbor search mechanism, establishing bidirectional correspondences between the source and target point sets. This overcomes the issue of traditional scale registration algorithms being prone to local optima due to unidirectional matching, significantly improving robustness in noisy environments. Second, the algorithm replaces the mean square error (MSE) with the maximum correntropy criterion (MCC), leveraging a Gaussian kernel function to adaptively weight matching point pairs and suppress outlier interference, further enhancing registration accuracy and robustness. Experimental results demonstrate that Bi-MCCSICP outperforms traditional ICP and its variants in terms of registration accuracy and stability in challenging scenarios such as noise, low overlap, and scale variations.