Image Super-resolution Algorithm Based On Multi-feature Fusion Learning

The Image Super Resolution Reconstruction Algorithm(Image Super Resolution,SR)is a technology that reconstructs low resolution images(Low Resolution,LR)into rich high resolution images(High Resolution,HR)through computer software technology.The current image SR technology is mainly researched on the basis of interpolation,reconstruction and deep learning.However,as the experimental requirements for the quality of the reconstructed image become higher,the traditional methods have problems such as blurry and aliasing of the reconstructed image,which cannot meet the specific work requirements.Therefore,researchers have applied deep learning to the field of image SR reconstruction,and have made breakthrough progress.This thesis mainly completes the following work based on the deep learning :(1)Strengthening the connection between different image features: The image features are divided into three types: original features,shallow features and depth features.Traditional algorithms do not take into account the different importance of these three features,and lack of The connection between features.We establish an adaptive correction channel to improve the processing capabilities of different features.In this paper,different features are processed differently to obtain complete high-frequency information as much as possible,and feature fusion is used to improve the model's representation capability.(2)Solving the problem of loss of high-frequency components of the image:when training the model,we find that the high-frequency components of the imagewill be lost as the depth of the model increases.Therefore,the multi-path residual learning method is used to uniformly cascade the obtained image prediction information to the reconstruction layer.The reconstructed image details are more abundant than those obtained by other methods,and the peak signal-to-noise ratio and structural similarity are higher.The overall network can solve the gradient explosion and overfitting problems in the model through the recursive structure.The recursive layer shares parameters,which solves the problems of increased model depth and excessive control parameters.(3)Testing the model performance through a large number of experiments: In the test of the general test set and the actual image,the rich details of the reconstruction result are displayed subjectively.According to different image evaluation standards,it proves the superiority of the model in terms of objective data.Experiments show that the method proposed in this paper can restore finer texture and sharper edges,and can be applied to specific work,and has important application value.