The registration of point clouds is a fundamental task in computer vision with applications in 3D image retrieval, segmentation, and shape recognition. This paper addresses the challenges of point set registration, considering factors such as nonrigid spatial transformations, high dimensionality, noise, and outliers. The focus is on probability-based methods, specifically leveraging the Stochastic Block Model (SBM). The proposed approach involves representing point clouds as graphs and introducing a latent variable for clustering. The proposed approach involves representing point clouds through graphs, introducing a latent variable for clustering. To enhance sparsity and computational efficiency, the model incorporates a Zero-Inated Normal distribution, focusing solely on non-zero entries below a specified threshold. The paper outlines the joint distribution and presents a variational inference algorithm for parameter estimation. The methodology provides a probabilistic framework for robust point set registration, demonstrating its potential in complex scenarios with high-dimensional data. Being a working paper, the following steps will concern numerical examples on simulated and real dataset.