Super-resolution reconstruction method based on multi-path deep convolutional neural network

Abstract

The invention discloses a super-resolution reconstruction method based on a multi-path deep convolutional neural network. The super-resolution reconstruction method comprises the following steps: acquiring a total training set; performing image preprocessing on the total training set; preparing a test set; reconstructing an image by using a convolutional layer of the convolutional neural network.According to the multi-path convolutional neural network structure provided by the invention, multiple branches are added to an original single-path neural network, so that image features with different scales can be processed by convolution kernels with different amounts, and reconstruction quality and visual effects are both improved compared with those of an original method while total parameter quantities are not increased.

Description

Technical field: [0001] The invention relates to a super-resolution reconstruction method based on a multi-path deep convolutional neural network, which belongs to the technical field of image processing. Background technique: [0002] With the development of science and technology and the progress of human society, information exchange and processing are becoming more and more important. Compared with text, sound, smell and other information, image information has the characteristics of intuition, image and large amount of information. Studies have found that in human Among all the information obtained, the proportion of visual information is as high as 60%, so images have become an important source of information in people's work and life, and are of great significance to the research and processing of images. [0003] Resolution is a measure of the precision of a screen image, and it is the number of pixels displayed on a monitor. The points, lines, and planes of the scr...

AI Technical Summary

Problems solved by technology

[0008] Super-resolution reconstruction algorithms are usually divided into three categories: interpolation-based, reconstruction-based, and learning-based super-resolution reconstruction algorithms. Interpolation-based methods are simple and computationally efficient, but the reconstruction results are too smooth to produce high-frequency details; reconstruction-based The methods based on learning cannot deal with complex image structures in natural images; learning-based methods rely on sample image libraries to establish a mapping between low-resolution images and high-resolution images. When a new low-resolution image is input, according to this mapping relationship Reconstruct high-resolution images, so the key to this type of method is the learning algorithm itself for the sample image library. The present invention mainly uses a learning-based super-resolution reconstruction method

Images