Research On Technologies Of Texture-Based Chest DR Image Retrieval

With the development of medical equipment, and the application of direct digital X-ray photography system in the hospital, the conventional X-ray imaging has been gradually replaced by the DR imaging. As the lung diseases are still the common disease in hospitals, a lot of chest DR image information is generated. By use and retrieve these images information effectively, it can not only provide great help for research and education work, but also provide computer-aided diagnosis for the doctor determine the patient's lung disease.The principle and key technologies of content-based image retrieval are described comprehensive, including the parts of feature extraction, similarity matching, MPEG-7, and systematic evaluation, and a variety of texture feature extraction methods based on transformation and statistical are described. The chest DR images shot in the hospital include the neck, armpit, heart, lung, and other structures. In order to research the lung image accurately, firstly, four kind of images in the image database are selected according to disease, and then the interested parts of image are segmented using an automatic segmentation algorithm, finally, the images are made the image enhancement processing through histogram equalization.It uses the method of GLCM to extract the texture features in the paper, and calculate the six figure characteristics of the angular second moment matrix, contrast, correlation, entropy, the deficit moments, and variance as the feature vector, then seek similar distance by calculating the Euclidean distance. It uses wavelet transform to decompose the image, by calculating the energy of sub-band. An improved image retrieval algorithm based on wavelet transform is proposed. Firstly, an image is decomposed by binary wavelet and the coefficients of each component are extracted, secondly, row modulus maxima density, column modulus maxima density and modulus maxima energy of each component are computed by analyzing the coefficients, and the eigenvector of the image is composed by them, finally, the similarity between the target image and each image of the database is measured, and according to the order of the similarities, the retrieval results are displayed. The methods to extract texture features based on multi-band wavelet transform and wavelet packet transform are presented.The retrieval results of four kinds of DR image based on the GLCM, two-layer wavelet transform, three-layer wavelet transform and improved algorithm are got in the experiment. It gets that the retrieval rate of improved image retrieval algorithm based on wavelet transform is higher than others through a large number of experiments, and it gets that the improved image retrieval algorithm based on multi-band wavelet transform and wavelet packet transform are better than the algorithm based on the binary wavelet transform.