Research Of Super Resolution Reconstruction Algorithm Of Medical Images Based On Sparse Representation

With the popular development and application of digital medical imaging technology in medical field,a variety of medical image imaging and processing systems that have been widely used in medical imaging image storage,image information retrieval,clinical therapeutic application,scientific experiment exploration and remote diagnosis of communication system,which will play an increasingly important role.Due to the limitation of sampling time,noise pollution and other factors,the acquired sampling data are often incomplete or impaired,which limits the analysis and understanding of medical images.Super-resolution reconstruction technology can improve the quality of degraded images and improve the resolution,which has a good application prospect and important scientific significance to medical image.Aiming at the above problems,in order to improve the resolution of medical images,this paper focuses on the study of sparse representation and dictionary learning.The contents and contributions of this dissertation are as follows.This paper summarizes the current status and development trends of super-resolution image at home and abroad.In view of the challenges of the current super resolution technology,the paper deeply studies and develops the solution based on the new technology.In this paper,the super-resolution reconstruction technology is used to post-process the single frame medical image,which is mainly from the following two aspects: first,a non-negative sparse representation model where dissimilarity of image patches used as a new penalty function instead of the similarity between image patches is integrated,which is proposed.It is introduced as a regular term in the sparse weight model,which can improve stability of the sparse coefficient and accuracy of the training dictionary,so that the input low-resolution image patches and examples of high/low-resolution image patches pairs in the database to achieve a higher degree of matching,and then restore the required high-resolution images patches,after experimental verification,this algorithm can improve the resolution and can effectively suppress the noise;second,on this basis,to improve the time performance of the algorithm is extremely important in the application of medical images,in order to improve the performance of the algorithm,the high-frequency information(HF)will be estimated is considered as a combination of two components: main high-frequency(MHF)and residual high-frequency(RHF),this paper proposed a medical image super-resolution usingdual-dictionary learning and sparse representation,Firstly,the observed images are divided into main components and redundant components,and then their main and redundant dictionaries are trained separately.The main high-frequency details and redundant high-frequency details are restored,and finally,the super-resolution images are recovered,which makes of the main dictionary and the residual dictionary learning recovering the MHF and RHF,respectively.Experimental results on test image show that by performing the proposed two-layer progressive method,more image details can be recovered and much better results can be achieved than that of existing methods,but also provide reference for medical diagnosis.