Research On Image Super Resolution Based On Feature Learning

With the rapid development of computer technology and Internet technology,the emergence of redundant multimedia information(such as video,image,sound,etc.)has appeared on the network.These complicated and diverse data bring human into the era of big data in computer network.The number of such information is large and stands for more memory space,which is not easy to store and transport.How to sort out these complicated information from the Internet for effective retrieval,preservation,mining which contains the value of technology,business and life,which become the urgent problems to deal with for human.Human beings acquire information through the senses of the facial features.And the main means for human beings to receive and perceive the surroundings are visual and sounding.According to the latest statistics,the information obtained by human vision accounts for nearly eighty percent of all information[1]obtained by human.As the saying goes: the eyes of the human mind is the window.The human eye sees a large amount and wide range capacity of information.For the sampling and recovery of information,the traditional Nyquist sampling principle [2] gives a general framework and defines the conditions that can accurately restore the original information,that is,the sampling frequency should be greater than twice the signal bandwidth.Due to the limitation of the principle and the characteristics of the information itself,the difficulties in the storage and transmission of information are increasing rapidly.In recent years,the emergence of Sparse Representation theory has attracted special attention from researchers and scholars.This theory avoids the limitation of the traditional Nyquist sampling principle,projects the signal into a specific transform domain.And according to the signal in the domain of the unique sparse features and optimization methods,the purpose of recovering the original signal can b achieved.In recent years,machine learning,artificial intelligence and pattern recognition technology attracts people’s attention.Based on the principle of sparse representation,this paper proposes a super-resolution reconstruction algorithm based on image feature learning,which integrates of machine learning,artificial intelligence and other related fields of knowledge.The typical feature extraction algorithm extracts a large feature length,and the algorithm takes up more space in the run-time,resulting in a high computational complexity.In the stage of feature extraction of the dictionary training process,this paper reduces the computation time and improves the efficiency of the algorithm by extracting the middle frequency characteristics of the image.And the popular K-SVD method with high efficiency is used to train the dictionary.The PCA(Principal Component Analysis,PCA)[3] method is used to reduce the dimension of the feature block before the dictionary training to further reduce the complexity of the algorithm.