Feature-based registration methods are often used in non-rigid image registration,among which joint estimation method based on point features is the most commonly used.The traditional joint estimation method based on the iterative closest point is easy to fall into the local optimum.To address this problem,in this paper,a fuzzy assignment algorithm is proposed,with which the distance between point pairs is used to generate the corre-sponding matrix,and the value of this matrix changes continuously with the change of the relative positions between point sets,replacing the 0-1 value pattern of the corresponding matrix in the traditional algorithm.In this way,the potential corresponding points can avoid being eliminated prematurely,and the convergence to the global opti-mum can be achieved with maximum probability.At the same time,the distance control coefficient is gradually re-duced in the iterative registration process to make the registration process from being coarse to fine.The simulation results show that,compared with shape context,center prealignment and traditional methods,the registration errors of this method can be reduced to 77.5％,32.6％and 23.2％respectively.
维普
万方数据