Improved image reconstruction of multi-scale residual network

Abstract:
Since the existing image reconstruction algorithms using convolutional neural networks cannot extract more feature information well, and the reconstructed image is prone to the problems of edge sawtooth effect and detail blur. Therefore, an image super-resolution reconstruction method based on multi-scale residual network is brought up. Firstly, the different scale features of the input image is abstracted by the multi-scale feature module. Secondly, the deeper image information is further extracted by the residual dense module, and the cross-channel interaction is realized by combining the onedimensional convolution attention mechanism to pay attention to the crucial information. Finally, the image reconstruction is completed by sub-pixel convolution.