Research On Key Technology Of Image Super Resolution For Full Scale Multiplication Ratio

In current society,image is one of the most important carriers of information in human daily work life.With the development of science and technology,the need of image quality is getting higher and higher.At present,in many application fields such as medical,transportation,aviation and other fields,there are very high requirements for image quality.High-quality images contain a lot of details in content,so it has become one of the current research hotspots for improving image resolution.In terms of hardware for obtaining images,research on obtaining high-quality images by hardware has reached a bottleneck at the technical level due to volume and cost,so from the software point of view,super-resolution processing of images has emerged as a key technology.Super-resolution processing of images has always been a hot topic at home and abroad,but for super-resolution studies of arbitrary multiplication ratios,there are still few corresponding results.At high multiplication ratios,the image sharpness becomes lower,and it is difficult for the naked eye to recognize the content of the image.With the advancement of technology,machine vision has begun to recognize images with low definition,and super-resolution technology for any multiplication ratio is studied.It has become increasingly important.In this thesis,a threshold adaptive method is proposed for image processing in the full-scale arbitrary multiplication ratio.The main research results are as follows:(1)The existing image processing algorithms are summarized and classified according to different standards.The principles of the four classical algorithms with different principles are elaborated and studied,and the corresponding simulation experiments are carried out.Then the four algorithms are compared and analyzed.(2)For the learning-based super-reconstruction algorithm,this thesis firstly introduces two kinds of dictionary training methods SVD and K-SVD,and on this basis,the CK-SVD algorithm is proposed.By selecting the appropriate threshold,the absolute product between the adjacent atoms is compared.The proposed algorithm deletes the atoms in the dictionary that have little utilization,and leaves only one of the atoms,and replaces the other atoms in the image block with the signal atoms that represent the least.This method can greatly reduce the computational complexity of the entire training process and reduce the time of image processing.Then by combining the CK-SVD algorithm with the ANR algorithm which has the best processing effect,the ANR-time algorithm is further proposed.Experiment results show that the proposed algorithm has a consistent improvement in running time as compared with other algorithms under both low multiplication ratio and high multiplication ratio.(3)For the arbitrary multiplication ratio processing of images,the image processing in the domain of arbitrary multiplication ratio is carried out for five classical algorithms,and two classical image processing mechanisms are introduced.Then according to the image processing results,in order to improve the processing efficiency of the image at any multiplication ratio,a full-scale concept is proposed,which provides a theoretical basis for the following algorithms.(4)For the content of full-scale arbitrary multiplication ratio image processing,this thesis proposes the scale-adaptive reconstruction(SAR)algorithm.Firstly,the priority criterion is proposed for the selection of image reconstruction algorithms.According to the sample processing results,several optimal algorithms are selected by using the priority criterion,and these selected algorithms are directly applied to the entire image set.Then,the sum Q of the full-scale peak signalto-noise ratio is proposed,and the slope ? of the Q value under any multiplication ratio is calculated,and the range of the threshold q is determined by comparing the magnitude of the slope ?.Finally,based on the sum of binary mode(SBP),the SBP(q,i,n)values of different images in different subsets of samples are calculated and compared,and the final switching threshold q is finally determined.The experimental simulation is carried out according to the proposed SAR algorithm.In the range of full-scale arbitrary multiplication ratio,the same image set is used to compare the experimental results of different algorithms.The results show that the proposed algorithm has wide applicability and has a significant improvement in the whole multiplication-ratio range.