Image method based on deep layer residual error CNN, device and storage medium

Abstract

The embodiment of the invention provides an image super-resolution method based on deep layer residual error CNN. The image super-resolution method comprises the steps that deep layer residual error CNN models are formed, low-resolution images are input into the deep layer residual error CNN models to be recognized, recognized images are obtained; and the recognized images are input into VGG lossnetworks to carry out recognition effectiveness evaluation, and recognized images with super-resolution are obtained. The invention further provides an active interaction device and non-transient state readable storage medium to realize the image super-resolution method based on the deep layer residual error CNN. The image super-resolution method based on the deep layer residual error CNN can enables the recognized images to have the advantages of high-frequency information, sufficient and reasonable texture detail and lower noises.

Description

technical field [0001] Embodiments of the present invention relate to the technical field of deep learning, and in particular to an image super-resolution method, device and storage medium based on deep residual CNN. Background technique [0002] Image super-resolution (Super Resolution, SR) is the process of upgrading a low-resolution (Low Resolution, LR) image to a high-resolution (High Resolution, HR) image through a certain algorithm. Existing techniques propose super-resolution reconstruction based on perceptual loss functions, and successfully generate images of high perceptual quality with sharper edges. Although the performance of image reconstruction has been greatly improved, the texture details of the reconstructed image still need to be improved. At present, in terms of loss function, for a network whose output result is an image, the usual practice is to use the sum of the pixel-by-pixel Euclidean distance between the result and the true value as the loss funct...

AI Technical Summary

Problems solved by technology

Practice has proved that the training result of the Perceptual Loss function has a lot of high-frequency information. Although it can overcome the lack of high-frequency information in the pixel-by-pixel loss, it will generate more noise in the texture area.

Images