Medical Image Retrieval Algorithm Based On The Compressed / Uncompressed Domain

With the rapid increase of medical image amount, the research of CBIR (Content-based image retrieval) and its application on CAD (Computer-Assistant Diagnosis) receive more and more concerns. The thesis researches and explores the algorithms for medical images' visual feature extraction and retrieval methods based on compressed domain and uncompressed domain respectively.The thesis firstly introduces CBIR's concepts, architectures and then explores a quantitative algorithm to extract visual feature within HSV color space according to the statistical histogram. In order to compensate the lack of spatial information of color features and make use of the rich textural information, a textural modal based on color co-occurrence matrix is presented consequently. To improve the system's learning ability, an algorithm by fusing visual features of color, texture and shape comprehensively and relevance feedback technique is presented, which is applicable to endoscopic images.In order to improve CBIR's retrieval speed and efficiency of retrieving huge amount of clinical JPEG images in compressed domain, the thesis briefly introduces JPEG compression standard and DICOM image standard compressed by JPEG, and then explores a feature extraction way by using specified DCT coefficients histogram without decoding fully.Based on the research of substantial algorithms described above, experimental prototypes based on compressed domain and uncompressed domain are developed respectively and the experiment results demonstrate the feasibility and efficiency of these algorithms.The related results in this thesis can be widely applied to clinical images management systems, such as PACS, HIS, RIS and so on.