关键词: 图像配准, 互信息, 粒子群优化, Powell法

Abstract: The registration method based on mutual information has advantages of high precision，excellent robustness，etc.But there are many local minimums in the registration function of mutual information，which put difficulty on image registration.In view of this problem，the paper proposes the normalized mutual information as similar estimation.In the process of image registration，a low precision solution is solved by Quantum Particle Swarm Optimization（QPSO） algorithm firstly and then a high precision solution is acquired by Powell method.The QPSO algorithm has strong global search capability and Powell method has strong local search capability.Experimental results show that the proposed hybrid method can overcome the problem of the local minimums in the mutual information function and initial point dependence problem in Powell method.The hybrid method has improved the success rate of image registration，and has high registration precision and high speed.

Key words: image registration, mutual information, Particle Swarm Optimization（PSO）, Powell method