Brain Magnetic Resonance Image Registration Algorithms Based On Hybrid Corner Detection

With the development of imaging technology, the modality medical images havebeen used widely in clinical diagnoses and surgical therapies.Integrating those images ishelpful to improve the accuracy of clinical diagnoses and surgical therapies.Imageregistration is the key part of integration;The research on brain magnetic resonance image registration based on hybridcorner detection is deeply studied in this paper under the support of Central CollegeFund. During the project, the image registration is converted into an optimizationproblem with medical background. The author studied and improved corner operators,corner matching algorithms and dynamic optimization manner so as to improve theaccuracy of image registration. Based on the study, the author gave a comparison ofbrain magnetic resonance image registration performance based on two cornerdectection operators, thereby proposing one kind of hybrid corner detection algorithmfor brain magnetic resonance image registration, and realizing dynamic brain magneticresonance image registration based on PSO algorithm based on inheritance idea. Theauthor also tested the algorithms using lots of brain magnetic resonance images underMATLAB platform. The results show that the algorithms proposed in this paper canachieve higher registration accuracy and better stability.The major contributions of the thesis are as follows:Compared the brain magnetic resonance image registration performance based ontwo corner detection algorithms, which include Harris operator and SUSAN operator.This part compared the performance of two corner operators, and analyzed theirdifferences and characteristics. This part gave great help for the design of cornerdetection algorithm in brain magnetic resonance image registration.Since the existing corner detection algorithms dosen’t considering syntheticallyaccuracy, distribution, strength of corners, they can not obtain the higher registrationaccuracy. The author proposed one kind of hybrid corner detection algorithm for brainmagnetic resonance image registration. This method merges the corners which areextracted by Harris and Susan operators. Then, this method conducts weightedcomputation based on two weight coefficients1and2. After that, the corners canbe chosen further. Through bidirectional maximal normalization relevance matchingalgorithm and voting mechanism, the final corners are chosen further and are matched between reference image and image needing registration. Finally, the Powell algorithmis used and the final transform coefficients are obtained.Realized dynamic brain magnetic resonance image registration based on hybridcorner detection algorithm. This algorithm adopted PSO algorithm with inheritance ofexcellent population. This algorithm inherited and changed the best population fromcurrent image registered, and then used it to guide subsequent image registering. Thenew method optimized initial population, overcame the flaw of too long time whenregistering many images continuously based on traditional manner which generatesinitial population randomly. The experimental results show that the new method canobtain higher registration accuracy and achieve stable image registering compared withtraditional method..The structure of this paper is as follows: chapter one illustrates the meaning andobjective of the project, and relevant the current research situation; chapter twoillustrates theoretical basis of image registration based on feature points, including basicconcept、basic procedures and image registration based on corners; chapter threeconducts the study on hybrid corner detection algorithm for brain magnetic resonanceimage registration: including relevant theoretical analysis, the realization of thealgorithm and experimental results; chapter four conducts study on mechanism analysisof hybrid corner detection algorithm: including weight coefficients and the image ofcorner mathing; chapter five makes study on dynamic image registering for brainmagnetic resonance image registration based on hybrid corner detection, includingrelevant theoretical analysis, the realization of the algorithm and experimental results;chapter six illustrates the major contribution of the paper and the future work.