Research Of Sparse Representation Super-Resolution Technology Based On Classified Dictionaries

High-resolution images mean high-density pixels,which can provide more details,not only can better meet the needs of computer applications,but also better meet people's visual sensory needs.In practical applications,we can improve the resolution of the image in both hardware and software.Hardware-based super-resolution methods are often costly,and in many cases its resolution is still difficult to meet the actual needs.Therefore,to improve image resolution by software programming becomes very meaningful,which not only reduce the cost,but also provides more details of the information.This paper mainly studies the super-resolution image reconstruction algorithm based on the classification dictionary.The main research contents and contributions include:Firstly,the principle and application field of super-resolution reconstruction technology are expounded,and the mainstream super-resolution reconstruction technology is analyzed and compared.The main problems of the super-resolution are analyzed and elaborated,and the evaluation standard of reconstructed image is expounded.Considering that the single dictionary can not express the image well,this paper presents a classification dictionary algorithm based on structural tensor.Structural tensor is widely used in texture analysis,corner detection,and optical flow estimation due to the estimation of direction and analysis of image structures.Therefore,this algorithm constructs a descriptive vector based on structural tensor to characterize the local features of the image block.The texture information based on the structural tensor of the image block is used as a priori and the guided image block is clustered so that the image blocks belonging to the same class have similar patterns,and the trained sub-dictionaries are more specific.The simulation results show that the classification dictionary obtained by structural tensor is better than the image reconstruction of a single dictionary under the same experimental conditions.Aiming at the problem that the time cost and clustering number of clustering algorithm are difficult to be determined in the above method,this paper proposes a classification dictionary algorithm based on image hash.The algorithm first extracts the gradient feature of the image block and obtains the hash value through the hash function.Then the image with the same hash number is divided into the same image bucket,which achieves the effect of image classification.This method not only improves the speed of image classification,but also makes the image structure more detailed.The algorithm obtains the corresponding sub-dictionary by performing separate dictionary training on the image blocks in each bucket.Experiments show that the algorithm has improved both subjective and objective,and the speed of reconstruction is improved obviously.