With medical imaging processes rapidly advancing, medical imagingtechnology act as an increasingly important role. In clinical nerve fibers study,orientation, density and length of white matter fibers were the focus of research,and nerve fibers tracking provided reliable references. The quality of diffusiontensor images and registration performance influenced the accuracy of whitematter fiber tracking. Hence, either in clinical practice or in medical research,the quality of image registration is pivotal. This paper focused on theregistration algorithm based on tensor information for magnetic resonancediffusion tensor images; proposed a more reasonable simulation method ofmismatch image; improved the genetic optimizing algorithm to accelerateregistration.Conventional tensor image registration methods were studied. Earlierregistration methods mainly focused on mutual information method. Due totensor information was not in consideration, registration accuracy was lower.Tensor reorientation method involved two steps, lower registration efficiency.To improve it, tensor overlap based registration method was proposed in thispaper.Three image interpolation methods after geometric transformation werediscussed and compared. Experiment results based on clinical data showed thatpartial volume interpolation method eliminated multiple local minima ofsimilarity measure.A hybrid optimization algorithm, consisted of genetic algorithm and Powellalgorithm, was proposed. Simple Powell algorithm converged to local minima,and conventional genetic algorithm took longer time to registration. The hybridoptimization algorithm ensured the accuracy in an acceptable registrationperiod. |