Medical Image Retrieval With Diffusion On TPG And Similarity Of Textons

With the extensive application of medical imaging equipments and rapid development of computer technology,a large number of medical image data are being produced everyday in the clinic.However,it become an urgent problem which cries for solved that how to manage these image data effectively and then reasonably apply them to the clinic diagnosis process.Traditional medical image retrieval has many problems,such as totally dependence on artificial recognition,limited to text retrieval,the low efficiency of image call.In this case,Content-based Medical Image Retrieval technology has developed rapidly in the field of medical image processing.It overcomes the limitation of traditional retrieval methods by combining the image perception,pattern recognition and so on.This paper focuses on the research of image feature extraction and image feature similarity measurement based on CBMIR technology.It introduce the current research algorithms of image retrieval,and systematically describe the common methods of feature extraction,similarity measurement and performance evaluation in medical image retrieval.Considering the special characters of medical images,the textons are used to extract local feature of the image.When comparing the features,the similarity between the images sometimes do not match the distance which reflects in Euclidean space.Thus we utilize a graph diffusion process to propagate the similarity information on a tensor product graph,the similarity between images are reevaluated by the global information.At first,we propose a texture feature extraction method based on textons to extract useful feature information of medical image.The texture is considered as a series of textons which are distributed according to certain rules,and described by its statistical properties.The similarity between the query and the database images is obtained by comparing the texture information between them.The algorithm which directly retrieve image by similarity measure have over reliance on the underlying features,the difficulty of its similarity to the expression and the effects of noise.For improving the retrieval efficiency,a diffusion-based approach on a tensor product graph was proposed to improve the texton-based pairwise similarity metric by context information of other database objects.Firstly,medical image features were extracted by texton-based statistical methods,and then the pairwise similarities were obtained with weights determined by the similarities between textons,a global similarity metric was achieved by utilizing the tensor product graph to propagate the similarity information along the intrinsic structure of the data manifold.Experimental results of ImageCLEFmed 2009 standard database show that,compared with the traditional algorithms,the proposed algorithm can suppress noise and reflect the similarity between images more objectively and truly,and futher improve the accuracy of image retrieval.