Research On Image Super-resolution Reconstruction Technology Based On Learning

With the rapid development of computer technology and information technology,digital images are widely applied in various fields of life.However,due to the cost of hardware and other reasons,the image or video obtained may be after the noise pollution of image and video with the low visual quality.In particular,the era of ultra high definition resolution display has begun,people have a higher demand for image or video resolution.Therefore,processing the acquired images or videos in the later stage by software is becoming one of the most attractive research fields in the field of image processing.To solve these problems,we studied the technology of image super-resolution based on learning.From the aspects of preprocessing,image reconstruction and image denoising,we designed the algorithm of image super-resolution reconstruction.The main contents of this paper are as follows:Firstly,the main idea of local neighborhood embedding algorithm is elaborated,and the principle of traditional neighborhood embedding image reconstruction algorithm,the steps of sample training and image reconstruction process are introduced,and the shortcomings of the algorithm are analyzed.Secondly,a pre-amplifier non negative constraint face image neighborhood embedding super-resolution reconstruction is proposed: in the training of high and low resolution image,the amplification of low resolution images,which have more similar manifold structure between the high and low resolution image;relaxing the constraint on the reconstruction of the coefficient of the iterative update block to obtain the reconstruction weights of high resolution images,resulting in better performance than the traditional algorithm of neighborhood embedding.The experimental results are superior to the traditional algorithm in subjective visual effect and objective evaluation,which proves the effectiveness of the algorithm compared with the traditional algorithm.Thirdly,based on the super-resolution reconstruction algorithm of sparse representation,an improved online dictionary learning algorithm for image super resolution reconstruction is proposed: in the sparse reconstruction dictionary training stage,online dictionary learning is used to get the best over complete dictionary.In the sparse representation stage,considering the redundant information between images,we reconstruct the high-frequency components of target blocks from similar samples,construct L1 norm regularization compensation pairs,suppress the noise in the sparse representation process,and improve the reconstruction effect.Experiments show that the algorithm can better restore the details of the image,and can effectively suppress the influence of noise,and improve the visual effect of the reconstructed image.Again,in the realization of the image super resolution reconstruction system,this paper designs an system in the GUIDE development environment of MATLAB.The system supports local image import,providing multiple super resolution reconstruction methods,supporting image contrast before and after reconstruction,and preserving super resolution reconstruction pictures to local hard disk.Finally,the paper summarizes the work,analyzes the shortcomings of the proposed algorithm in this paper,and the key issues in the super-resolution reconstruction problem,and prospects the future research.