Research On Techniques Of Medical Image Retrieval Based On Feature Spatial Distributing

With the development of medical imaging techniques, lots of medical images are generated in hospitals. However, it is very hard to find images meeting our need from the large number of images. As a result, it is becoming more and more important to manage medical images efficiently using computer. Content based image retrieval (CBIR) makes use of low-level image features for example color, texture and shape to express images and performs well. So that using CBIR technology in medical images is becoming a hot research area.The extraction and expression of image feature is the base of content-based image retrieval (CBIR). So, how to extract low level features which reflect the high level semantics of an image is crucial for medical image retrieval. In this paper, after knowing the methods of image feature extraction the feature extraction algorithms that suits to medical are researched importantly. We make a discussion of color, texture, shape and other image low-level features. In addition, we do some study on evaluation standard of retrieval and relevance feedback.In this thesis, a feature extraction method based on valid dimension selection is proposed for medical images. We use image-block to extract feature from images, so that the local information of images is captured. Then an unsupervised cluster is implemented on dimensions to find valid dimensions. At last, the valid dimensions are used for medical image retrieval. The experimental results show that the approach proposed can reduce dimensions effectively and a preferably performances of retrieving is achieved. On the other, a local similarity retrieval method is proposed based on block area choosing. The local area is obtained by block selection. This method can be used to find medical images that are similar in local area. In the end, we design and accomplish a CBIR system, and our experiences of the two methods described above are carried by the system.