Research On Image Compressed Sensing Reconstruction Algorithm Based On Convolutional Neural Networ

Image is one of the important carriers in Internet information transmission,therefore the resource consumption in its storage and transmission are the important problems faced by modern big data era.Image compression sensing technology combines compression and sampling processes,which can restore the original image at a lower sampling rate.This technology has the advantages of increasing compression rate and saving both storage resources and transmission bandwidth.So it attracts a lot of attention.However,the traditional image reconstruction algorithm has some problems such as long computation time and poor reconstruction quality.With the development of deep learning in the field of image processing,the combination of image compression sensing technology and deep learning algorithm provides great help for the breakthrough of image reconstruction.However,there are still some shortcomings in image reconstruction algorithms based on deep learning,and there is still room for improvement in image quality.Therefore,this thesis studies the image reconstruction algorithm based on deep learning,and the specific contents are as follows:(1)Aim at the problem of detail loss of reconstructed image,an image compression sensing reconstruction algorithm based on multi-scale feature fusion is proposed.In this thesis,we combine multi-scale residual connection and dense connection to enable the network to extract enough rich image features.To make full use of network layered features and improve the accuracy of image reconstruction,a dual attention fusion module is designed to integrate all local features,make the global selective aggregation semantics and weight the importance of each channel,and supplement the spatial information of shallow features for deep features,which is conducive to pixel positioning in the feature map.The experimental results show that compared with other algorithms,our algorithm significantly improves the image quality and gets more image details.(2)Aiming at the problem of block effect in the reconstructed image,an image-compressed sensing reconstruction algorithm based on the fully convolutional network is proposed.Convolution and deconvolution are used to replace the measurement matrix and full connection layer in the previous methods to avoid the destruction of image structure information due to image segmentation.The multi-scale residual connection and dual attention fusion module are added to the reconstruction network to strengthen the network’s learning of image texture region features,and the network’s shallow features and deep features are combined through the global dense connection to improve the network performance.Experimental results show that the proposed algorithm solves the block effect problem and has high image reconstruction quality.