The Study And Implementation For The Research On Feedback Technique Of Content-based Medical Image Retrieval

With the developing of digital image in the field of medical, medical imageequipments are emerging in an endless. Digital medical image is an important part ofmodern treatment, large number of medical images with pathology information aregenerated from hospital. The information and images are so important to clinical andmedical study. So, it has been a hot topic about how to search and save the magnanimityimage. The traditional image research is based on text annotations, that is keyword index.The advantage of this method is fast, accurate, but it is amount of work for labeling.Except for medical image, subjectivity and accuracy are not good for developing unit andstandardization. In order to solve the problem, the technology of Content Based MedicalImage Retrieval(CBMIR) appears, the technology is image retrieval based on content, it isaimed at the characteristics of image itself. Its characteristic vector is constructed by graylevel, texture, shape and other physicalcharacteristics. Feature matching and measurementalgorithms are selected to search and match object images and database images. Thepurpose of index is to get the maximum similar set between user feature set and objectiveimage, and the images are ordered as similarity descending. In this paper, after analyzingthe key process of searching, a kind of method, combined dynamic weight tuning withweighted markov distance algorithm, is proposed. The method combines five featurealgorithms of gray-scale concurrence matrix with two texture feature algorithms of tamurato construct rich compositive feature vector set. The improved weighted markov distancealgorithm is used as match algorithm and relevance feedback mechanism is set up, thefeedback information is used for adjusting comprehensive feature weights and modifyingsearch strategy. After several changes, the search results are more satisfying.The development platform of this study is VS2005, programming language is C++,database is SQL Server2005, accessory kits are OpenCV and ITK. The experimental results show that the method can improve feedback search, the recall ratio and precisionratio efficiently.