Research On Key Techniques Of Content-Based Medical Image Retrieval

In recent years, along with medical imaging technology in a wide range of applications has produced too large amount of medical technology images information,it becomes a problem that how to find needed image efficiently and rapidly in large-scale image database. Content-based image retrieval techniques have been applied in the field of medicine, in clinical, teaching, scientific research, as well as medical image archiving and communications systems (PACS) have an important role.How will the medical image retrieval and image combined for physicians to provide convenient and accurate image search tool for the diagnosis and the provision of complementary proposals, this article is a study of the target.This article focuses on medical image feature extraction, index technology and relevance feedback and so on, that such key technologies were studied.On the basis of the color histogram feature extraction technology, adaptive weighted method to improve the color histogram is presented, this method proved an important part of the image can enhance the features and much better for the calculation of similarity.Use binary information to indicate the image color, shape, texture characteristics, the features will be used to filter images that obviously does not meet the requirements, accelerate the retrieval speed of system; Use improved based on features of the weighted SOON clustering cluster image, on this basis, construct dynamic index tree for each category based on similarity hierarchical clustering, the index tree can be directly get the result that based on the similarity ranking, based on the SOON cluster image search index is 9.3 times that of the exhaustive search, and in this paper based on the similarity SOON clustering image index is 41.2 times that of the exhaustive search. Thus it other than the index tree has to find a faster efficiency;To unbalance of positive and negative sample, use relevance feedback based on BSVM methods, at the same time, for a small sample size and low accuracy, present improved way that combining priori knowledge of Boosting, given based on the combination of Boosting and BSVM relevant feedback method (BBSVM), on the base, in order to improve the speed and accuracy of image retrieval, present the long learning relevance feedback algorithm(LBBSVM). The results show that our proposed method of feedback-related LBBSVM feedback in terms of accuracy in regard to the speed of feedback is the best.Based on the above key technologies designed and implemented a content-based image retrieval system, the system achieve medical image feature extraction, image similarity search and display the images to the user, use relevant feedback technology to improve the accuracy of image retrieval. The results show that the system has good precision, recall, as well as efficiency.