Feature Extraction And Research Of Content Retrieval Based On Medical Grayscale Image

Medical imaging technology plays a vital role in the medical field because it not only provides important information about the internal organs of the body for clinical analysis and medical treatment but also assists doctors in the diagnosis and treatment of various diseases.Nowadays,medical image data is increasing at an explosive rate so that forming a huge medical image database,it is an extreme difficult task to retrieve similar medical images from such a large database.The retrieval of medical image content has been implemented by exploiting feature extraction,feature quantization and similarity metrics in this thesis.Image features are abstract expressions of image content,feature extraction of different image methods has different emphasis on image description.The ability to represent images could be enhanced by fusing multiple underlying features,especially for medical images that require more expressive features to describe.Feature extraction of medical image is the basis of image content retrieval,the extraction of visual features with strong expression ability and good anti-noise ability directly affects the accuracy of subsequent image retrieval.Improved algorithms of geometric corner features and fused texture feature of LBP have been implemented for fundus angiography images in order to obtain effective feature descriptors of images.At the same time,the multi-feature vectors of images are further quantified by combining deep learning in purpose of further abstracting image features,which not only makes up for the one-sidedness of the single feature description,but also improves the robustness of the feature descriptor.A comprehensive study on multi-feature extraction and feature fusion,similarity measure and feedback retrieval has been conducted based on the discussion and analysis of the research status of important technologies related to content-based medical image retrieval,according to the feature extraction of medical image.Different feature extraction methods are proposed based on two images of medical fundus image and medical X-ray image.In general,the main work of this thesis consists of:(1)The domestic and international development status about feature extraction of medical image has been investigated and the main techniques of image preprocessing,feature extraction and content retrieval have been summarized.(2)Achieved fundus image retrieval that combines corner features and texture features based on the present algorithm of corner feature extraction.(3)A retrieval feedback method based on updating the weight of features for medical fundus image has been proposed,which aims to improve the high-level semantic information of image retrieval.(4)The higher-level quantization of features has been implemented by exploiting linear coding and combining SURF and matrix features of gray level co-occurrence for X-ray medical images.The bag-of-features have been used to cluster features and optimize the mismatch in feature matching to improve retrieval efficiency during the process of image retrieval.