cGAN algorithm-based super-resolution image recovery technology

Abstract

The invention discloses a cGAN algorithm-based super-resolution image recovery technology, and relates to super-resolution recovery for deep convolutional antagonistic neural network images. The existing method needs to carry out a lot of training, the network layers are relatively shallow and the feature weight values included in models obtained through the training are not complete enough, so that the final recovery effect is not ideal. Greyscale maps are used during the training, so that the super-resolution recovery effect for three-channel colored images is not good. Aiming at the defects in the prior art, the invention provides a cGAN algorithm-based super-resolution image recovery method which is capable of training the colored images to achieve better recovery effect. The key points of the method comprises that (1) the training time is short, (2) three-channel colored maps can be directly trained, (3) much image preprocessing is not required, and (4) the models obtained through training can be used once for all.

Description

technical field [0001] The invention belongs to the field of computer image vision, and is based on generating deep convolution against neural network image super-resolution restoration. Background technique [0002] Since the birth of the AlexNet convolutional neural network in 2012, the neural network has become the mainstream of learning. Compared with the strong priori assumption (priori) of the Bayesian school, the repeated research of SVM on the kernel function (kernel), the neural network does not require researchers to pay too much attention to details, and only needs to provide a large amount of data and set hyperparameters , good results can be achieved. Journal literature (Goodfellow I J, Pougetabadie J, Mirza M, et al. Generative Adversarial Networks[J]. Advances in Neural Information Processing Systems, 2014, 3:2672-2680.) is the pioneering work of GAN. The principle of GAN is relatively simple. Here is an example of generating pictures. GAN includes two netwo...

AI Technical Summary

Problems solved by technology

[0004] In the method of Document 1, a lot of training is required, and the network layer is relatively shallow, so the trained model contains insufficient feature weights, so the final recovery effect is not ideal

Grayscale images are used for training, so the super-resolution recovery effect on three-channel color images is not good

Images