Research On Super-solution Reconstruction Of Medical Image Based On Multi-scale Clustering And Nonlocal Autoregressive Learning

Medical images have a wide range of applications in clinical therapies because it can provide the necessary information of patient organs that can lead to the precise treatment.Medical image data can truly reflect the shape of the patient's organs and the location of the lesion,which has important significance for guiding the patient's clinical treatment.However,the medical images obtained by the existing medical equipment are significantly influenced by the radiation dose,for example,the higher image resolution means that the patient needs to be exposed to more doses,which will undoubtedly have a bad effect on the patient.On the other hand,if the resolution is low,it is not easy for doctors to do diagnosis and treatment,which will have an adverse impact on the treatment of the patient.Image processing technology contains a lot of technology,and super-resolution reconstruction is one of the branches,which role is to improve the image resolution.The specific process is to find the additional priori information form a single or multiple low-resolution images,and by adding the priori knowledge to the image,the image resolution can be improved and the high-resolution images can be reconstructed.This technology is not only by the way to improve the hardware of the physical device but also by the method of processing an image with a computer after obtaining an image to improve the image quality.Besides,this is a post-processing method that improves image resolution.Based on the studying of the degradation model of the image and the characteristics of medical images,aiming at the existing problems of medical image,this paper applies super-resolution reconstruction techniques to medical images and proposes two super-resolution algorithm for medical image: medical image super-resolution reconstruction algorithm based on multi-scale clustering and medical image super-resolution reconstruction algorithm based on non-local autoregressive learning.Based on the super-resolution reconstruction algorithm of multi-scale clustering,the algorithm in this paper first uses the decomposition method of quad-tree to patch the image,so that the image blocks of different scales are obtained adaptively.Then,by analyzing thecharacteristics of medical images,combined with the characteristics of medical images,we extract the image feature and cluster on the image patch.The clustering centers with different scales of the corresponding image patches can be obtained.Finally,a simple function algorithm is used to learn the corresponding regression coefficients of each image patch,by which the high resolution image is obtained.Experiments show that this method can obtain satisfactory super-resolution reconstruction results both in terms of visual evaluation and quantitative evaluation.In the super-resolution reconstruction algorithm of medical image based on non-local autoregressive,we use the similarity of medical images' features to add the characteristics of medical image similarity to the autoregressive model.And then,this model is applied into the medical image super-resolution reconstruction model based on sparse representation to construct high-resolution medical images.Besides,in the process of dictionary training,the algorithm is not training the global dictionary but using the clustering algorithm to get classified sub-dictionary,so as to improve the efficiency of the experiment.The experimental result on the medical images indicates that the method in this paper can improve the resolution of the medical image.And since image got by our method is clearer,more details can be identified.What is more,the comparison between the noise ratio of peak signal and the similarity of the structure can also prove that our algorithm can improve the image quality.