Research On Techniques Of Texture Feature-based Medical Image Retrieval

Along with a variety of medical imaging techniques and equipment (such as CT, X-Ray, MRI) used broadly, a lot of medical images have been produced in hospitals. Medical image databases become important parts of computer aided diagnosis and prophylaxis medicine. It is very difficult that retrieving the given image from huge medical image database by manual work. So we need the exact and effective medical image retrieval by computer. Content-based image retrieval (CBIR) is a hotspot of image retrieval at computer filed. It is different from traditional textual retrieval. Thus it is very meaningful that it is applied at medical filed.Feature extraction and similarity measurement is the most crucial techniques in CBIR. So, how to extract low level features which reflect the high level semantics of an image and how to use these features to carry out similarity measurement is crucial for medical image retrieval. Otherwise, it is very important to optimize the result of retrieval using relevance feedback techniques.Based on analyzing the key techniques of CBIR systematically, we discuss the algorithms of texture features extraction based on co-occurrence matrix. In allusion to medical image retrieval, we present the algorithms of gray-cell difference co-occurrence matrix and gray-wavelet co-occurrence matrix to extract texture features. Then we compare these algorithms with gray co-occurrence matrix and gray-grads co-occurrence matrix and validate these algorithms for medical image retrieval using experiment. Finally, we design and implement a medical image retrieval system based on Euclidean distance.