A fast image super-resolution reconstruction method based on deep learning

Abstract

The invention relates to a fast image super-resolution reconstruction method based on deep learning, and belongs to the field of image processing. The method comprises the following steps of selectingan image training set and a test set, and performing feature extraction, nested network feature refinement and sub-pixel up-sampling operation on a low-resolution image by using a deep neural networkto obtain high-resolution detail residual information of the image; carrying out transposition convolution processing on the low-resolution image to obtain high-resolution space low-frequency featureinformation of the image; combining the high-resolution detail residual information of the image with the high-resolution space low-frequency characteristic information to obtain a high-resolution reconstruction result of image estimation; performing loss value measurement on the high-resolution reconstruction result of the image estimation and the high-resolution image block; updating the network weight by using an Adam operator to obtain a trained network model; and inputting a low-resolution image into the trained network model to obtain a high-resolution reconstructed image. According tothe method, the super-resolution reconstruction of the image is accelerated, and a good reconstruction effect is kept.

Description

technical field [0001] The invention relates to a fast image super-resolution reconstruction method based on deep learning, in particular to convolution network parameter adjustment in deep learning, loss function optimization, and an improved algorithm for the training process of deep network double-branch structure, belonging to the technical field of image processing . Background technique [0002] Single image super-resolution reconstruction technology (Single Image Super-Resolution, SISR) refers to a given single low-resolution image, through a specific algorithm to restore the details of the image lost due to downsampling, so as to obtain more detailed information , The process of high pixel density images with more delicate picture quality. Due to the ability to recover finer details with limited information, this technology has been widely used in various application scenarios, such as high-definition television broadcasting, video surveillance, and satellite imagin...

AI Technical Summary

Problems solved by technology

[0008] Step1, select the image training set and test set, and process to obtain low-resolution image blocks and high-resolution image blocks; in the step Step1, select the high-resolution image BSD200 data set that includes 200 png format and 91 bmp format for use The image set Train91 is used as the model training set, and Set5, Set14, BSD100, and Urban100 are selected as the standard test set; in order to solve the problem of excessive memory space load, when the magnification factor is 3, the image is cut into 48×48 low-resolution images Image block (Low Resolution, LR) and the corresponding 144×144 high-resolution image block y HR (High Resolution, HR);

Images