一种互信息与梯度信息结合的多模图像配准方法  

Multi-modality Image Registration Algorithm Combining Mutual Information and Gradient Information

作　　者：史聪文 赵勋杰[1] 

机构地区：[1]苏州大学物理与光电.能源学部,江苏苏州215006

出　　处：《光电技术应用》2016年第4期46-49,65,共5页Electro-Optic Technology Application

摘　　要：数字化X射线图像(digital radiography,DR)与数字重建放射图像(digitally reconstructed radiography,DR)属于不同模态图像,实现二者的高精度快速配准是一个技术难题。在实际应用中,往往会同时获取物体的正侧面DR和DRR图像。提出一种基于互信息与梯度信息相结合的配准算法。首先,对正侧面图像进行小波分解,获得低分辨率子图像并配准,使用粒子群优化(particle swarm optimization,PSO)算法进行全局寻优;然后,根据配准结果,判断互信息与梯度信息配准结果是否正确,如果配准错误,则在下一阶段中不使用该结果作为配准依据;最后,以PSO算法寻优结果作为Powell算法的寻优初始点,对原始正侧图像进行精确配准。实验结果显示,本算法快速完成配准,配准精度达到2 mm,满足实际应用要求。Digital Radiography（DR） image and digitally reconstructed radiography（DRR） image are differentmodality images, accurate and rapid multi-modality image registration is a technical difficulty. In practical applica-tions, front and side DR and DRR images are generally acquired at same time. A multi-modality image registrationalgorithm is proposed based on combining mutual information and gradient information. Firstly, the low resolutionimages, which are got by wavelet transform, are registered based on particle swarm optimization（PSO） algorithm.And then, we can determine if the registration results are true. If it is not true, we don＇t use it at the next process. Fi-nally, we set the results of last process as initial position of Powell and register original images. The experimental re-sults show that the algorithm can rapidly register multi-modality image, and the registration accuracy is 2 mm. Thealgorithm meets the practical application requirements.

关 键 词：图像配准 互信息 梯度信息 粒子群优化算法 POWELL算法 DR DRR 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]