Content-Based Medical Image Retrieval

Medical image retrieval tecnology has been widely used in the field of modern medicine andit plays an important role in teaching,clinical treatment, medical images’ filing and scientificresearch.In this paper, we first research on some image retrievals based on contour feature,andthen we employ image retrieval again by the wavelet transform.Through the study of medical image retrieval based on contour feature(see in chapter3),several algorithms for extracting contour feature are presented,then we find an optimal algorithmby setting a contrast experiment. In this paper,we use Canny edge detector to extract the contourfeature of image.First of all,we adopt the preprocessing to the initial image. In order to obtain agood effect on edge extraction and improve the algorithmic stability of edge detection, we use3×3median filter to denoise the image. For obtaining a better edge feature, we sharpen the imagebefore taking edge detection. In this paper,we use the Laplacian convolution to sharpen the image,adopt canny operator to extract the edge feature and Minkowski distance to measure thesimilarity. Therefore, the process of medical image preliminary retrieval are finished.The wavelet transform is widely used in the field of image processing, and we propose amethod of image retrieval based on the wavelet transform in the chapter4. The Mallat algorithmin our image processing refers to the2-D discrete wavelet transform and multi-resolution analysis.Hence, in this paper, we introduce the definition and the steps of Mallat algorithm in details.After we do the preliminary retrieval based on the contour feature, we retrieve the image for thesecond time by the method of the wavelet transform. The image is decomposed into three layers.The wavelet transform is of de-correlation property, however, this de-correlation property isincompletely.Therefore, after denoising the image by the wavelet transform, we adopt2-Dprincipal component analysis so as to remove the quality of correlation and reduce dimension. Bytaking2-D principal component analysis, the speed of retrieval is improved. At the same time, weuse the Minkowski distance to measure the similarity, then the second retrieval of the image iscompleted.Experiment results have showed that after the initial retrieval based on contour features, weuse the wavelet transform in order to do the second retrieval. Through the value of the recall ratioand the precision ratio, this mixed method has better retrieval effect than only use one way.