Image super-division method based on progressive residual generative adversarial network

Abstract

The invention discloses an image super-division method based on a progressive residual generative adversarial network. The method comprises the following steps: firstly, performing down-sampling processing on a given real target image to obtain a series of low-resolution picture sets; building a residual generative adversarial network, and starting layer-by-layer training from the lowest sub-network layer until a vivid reconstructed picture with the same resolution as the original real picture is generated; and finally, further performing up-sampling on the image according to a required super-resolution value, overlapping noise and inputting the noise into a higher layer of sub-network for iterative training, and finally obtaining a required super-resolution image. On the basis of the generative adversarial network model in deep learning, a weighted cascade residual structure is fused, feature information of the input picture can be efficiently and fully utilized, data features of the input picture can be fitted, and super-resolution of the picture is achieved. Compared with the prior art, the method is low in cost and easy to implement, and the super-resolution effect of the image can be improved.

Description

technical field [0001] The present invention mainly relates to the field of image processing, and specifically relates to an image super-resolution method based on progressive residual generation confrontation network. Background technique [0002] For a long time, image super-resolution reconstruction (referred to as image super-resolution) has been a branch of extensive research in the field of image processing. Specifically, image super-resolution technology is given one or a batch of low-resolution images, through hardware processing or algorithm processing, and under the premise of ensuring that the images are not distorted, a clear high-resolution image is obtained. This technology can enrich and improve the detailed information of the collected images, making the collected images more practical, so it has a wide range of application scenarios in aerospace, meteorological remote sensing, biomedicine and many other fields. [0003] Traditional super-resolution methods ...

AI Technical Summary

Problems solved by technology

Usually in the generator G network, the feature information extraction and fitting are directly completed through the multi-layer neural network, but this network form cannot make full use of the input feature information. If the Desnet form is used, the network structure is too complicated, which seriously slows down the network processing efficiency.

Images