Wasserstein distance-based fuzzy C-means clustering in Riemannian manifold feature space for image segmentation

Chengmao Wu,Jia Zheng

DOI: https://doi.org/10.1007/s11760-024-03705-0

IF: 1.583

2024-12-13

Signal Image and Video Processing

Abstract:To ensure the accuracy of image segmentation, selecting appropriate image features is crucial. The existing image segmentation methods mainly utilize spectral features of images to achieve image segmentation, which have low segmentation accuracy and weak robustness, and are difficult to adapt to the needs of complex image segmentation. Therefore, this paper proposes a new robust fuzzy clustering image segmentation based on Wasserstein distance in the Riemannian manifold feature space. At first, the spectral features of pixels in the neighborhood window around the current pixel are constructed into a Gaussian normal distribution structure model, and the original image is mapped to the Riemannian manifold feature space to achieve Riemannian manifold feature modeling of image feature information. Secondly, the Wasserstein distance is used to measure the difference between two Gaussian Riemannian manifolds, and a robust fuzzy clustering method for image segmentation is proposed in Riemannian manifold feature space. Finally, the local convergence of the algorithm is proved using the Zangwill theorem and bordered Hessian matrix. The experimental results demonstrate that the proposed algorithm has good segmentation performance and strong noise resistance. Compared with existing segmentation algorithms based on spectral feature space and Riemannian manifold feature space, this proposed algorithm is more effective and robust in segmenting images with or without noise.

engineering, electrical & electronic,imaging science & photographic technology