Hierarchical Representation Via Message Propagation for Robust Model Fitting

In this paper, we propose a novel hierarchical representation via message propagation (HRMP) method for robust model fitting, which simultaneously takes advantages of both the consensus analysis and preference to estimate parameters multiple instances from data corrupted by outliers, fitting. Instead analyzing information each point or hypothesis independently, formulate as alleviate sensitivity gross outliers. Specifically, firstly construct representation, consists layer layer. The is used remove insignificant hypotheses Then, based on an effective (HMP) algorithm improved affinity (IAP) prune vertices cluster remaining points, respectively. proposed HRMP can not only accurately number instances, but also handle multi-structural contaminated with large Experimental results synthetic real images show that significantly outperforms several state-of-the-art fitting methods in terms accuracy speed.