The Research Of Multi-feature-based Image Retrieval Of Medical Images

In the last half-century,there is a the rapid development of computer science and technology in the world, the digital image technology became a indispensable part of people’s daily lives. Medical imaging equipments using digital image have replaced the old machine in the medic.A high- level living standard makes people pay more attention to health care at the same time,and it leads to a increasing demand for med ical resources. Because it relates to the various principles of complex and diverse medical devices and the biological complexity of human’body, The large number of digital medical images, high professionalism make it’s hard to deal with these medical digital images.This thesis based on the sticking point of CBMIR, try to find a better way to retrieval of medical images,it analyses the medical image enhancement, feature extraction and fuzzy matching detailedly,make a medical image retrieval method with feasibility and effectiveness.The main works as followed.(1) This thesis analyses the nature and characteristics of medical digital image,using the theory of histogram of homogenization and median filtering to solve the problem about blurred image and high noise in CT images.(2) Medical digital image always have a high- level information complexity,and it makes a interference in feature extraction. Based on the theory which is using different features to make a optimization of medical image feature extraction,it talks over three type of feature extraction algorithm.Firstly,we try to use the gray- level histogram for extracting color feature from medical digital images,and find a way to deside how to divice intervals. Secondly, we try to use GLCM algorithm for extracting texture feature from medical digital images,and disscuss the standard of choosing advanced features,using and integrate GLCM features calculated under different conditions by weighted fusion method for increasing the information that extract form medical digital images.Thirdly, we try to use invariant moments for extracting shape feature from medical digital images,and work out different moments how to influenced by noise of images,then find a optimization for making moments to handle more easilier.(3) Based on the fuzzy theory,we find a way to transform a normal content feature of medical image into a fuzzy feature,and give a algorithm which is using these fuzzy features to calculate the similarity between two medical digital images by multi- feature fusion,improve the property of the retrieval system.We analysis the standard of choosing the membership function and the different ways to creat a membership function,and choose a advanced gaussian function as a effective membership function in medical digital image at the last.