A cluster network super-resolution image reconstruction method of an Laplace pyramid structure

Abstract

The invention relates to a cluster network super-resolution image reconstruction method of an Laplace pyramid structure. By dividing the images to be reconstructed into training data sets and test data sets and enhancing and preprocessing them respectively, a Laplace pyramid cluster network is constructed and the training data sets are input for training, and the test data sets are input into thetrained Laplace pyramid cluster network to reconstruct super-resolution images step by step. A Laplace pyramid structure is adopted, By progressively reconstructing high-resolution images, Step by step optimize reconstruction results, better solve super-resolution image reconstruction, the residual learning is applied to the network, Reducing network parameters and avoiding gradient explosion caused by network complexity increase. In each construction module, there are both forward and feedback connections between any two convolution layers, and the information between layers is updated alternately, which maximizes the information flow and feedback mechanism between layers, makes the connection between layers more dense and extracts more detailed features.

Description

technical field [0001] The present invention relates to the technical field of graphic image conversion in the image plane, for example, to create a different image from bitmap to bitmap, and in particular to a Lapp that can convert an existing low-resolution image into a high-resolution image Clique Network Super-resolution Image Reconstruction Method Based on Las Pyramid Structure. Background technique [0002] Super-resolution image reconstruction refers to the technology of converting existing low-resolution (Low-resolution, LR) images into high-resolution (High-resolution, HR) images by means of signal processing and image processing, which is suitable for Solve various tasks in computer vision, such as remote sensing satellites, security and surveillance imaging, medical imaging, image generation, etc. [0003] Common super-resolution image reconstruction methods include interpolation-based methods, reconstruction-based methods, and learning-based methods. The interp...

AI Technical Summary

Problems solved by technology

[0006] The paper "Image Super-Resolution Using Deep Convolutional Networks" first proposed the SRCNN method to learn nonlinear LR-to-HR mapping, and introduced a three-layer convolutional neural network into the super-resolution network; the paper "Accurate Image Super-Resolution Using Very The VDSR method proposed by "Deep Convolutional Networks" and the DRCN method proposed by the paper "Deeply-Recursive Convolutional Network for Image Super-Resolution" increase the network depth by clipping gradients, skip connections or recursive learning to reduce the training problems caused by the increase in network depth; the above Although the method is superior to the traditional learning-based method and is more suitable for the super-resolution reconstruction of general images, before the prediction, directly using the predefined upsampling operator to increase the input image to the desired output spatial resolution will Adds unnecessary computational cost and often leads to visible reconstruction artifacts

[0007] In order to reduce unnecessary computational costs and weaken visible artifacts, the paper "Accelerating the Super-Resolution Convolutional Neural Network" proposed the FSRCNN method, which no longer uses the predefined upsampling operator, and instead directly converts the original low-resolution image As input and introduce transposed convolution for upsampling in the last step; the paper "Deep Laplacian Pyramid Networks for Fast and Accurate Super-Resolution" proposes a super-resolution method LapSRN based on a cascade structure, which can generate at multiple resolutions Intermediate SR prediction and gradual optimization have achieved good reconstruction results; however, these methods all use a chained method to stack building blocks, and only pay attention to the forward information transmission between convolutional layers, ignoring the information between convolutional layers. Feedback and exchange of information

Images