| Since 1990s, as a great deal of image data having been produced, it becomes an urgent issue how to find out an interested image quickly from an image database. Thus the Content-based image retrieval (CBIR) technology emerged as the times require. Recently, there are many new CBIR systems emerging at home and abroad, which use the low level features of image such as color, texture and shape to retrieve images. CBIR is studied and tried to apply on medical images in this thesis. The experimential results show the retrieval performance is high. The latest Digital Imaging and Communication in Medicine (DICOM 3.0) is introduced in this thesis. The color descriptor, texture descriptor and shape descriptor are also introduced and compared systematically, and the related part of MPEG-7 is studied.The color pair algorithm is improved in this paper, so not only the stability of the algorithm is added, but also it is suit for all kind of different-size images and the retrieval performance is promoted greatly. According to characteristics of medical images, texture and shape features are used respectively and synthetically to retrieve medical images.The gray differential statistics is utilized to achieve texture-based image retrieval. It comes to a conclusion that if the differential value is between the threshold TO and T\, the value represents the texture of an object. On the other hand, if the differential value is larger than threshold T\, it represents the border of an object. Hu moments are also used to realize shape-based image retrieval in this paper. The four out of the seven Hu moments and the differential value larger than T1 are combined to achieve medical image retrieval. These improvements promote the performance of medical image retrieval significantly. In the end the texture and shape features are combined to complete the medical retrieval task. It is proved that the texture features and shape features are supplemental each other to medical images. |
PDF Full Text Request