Image super-resolution method based on cascade network framework and cascade network

Abstract

The invention relates to an image super-resolution method based on a cascade network framework and a cascade network, and the method comprises the steps: obtaining a preliminary upsampling result of atarget image based on a basic super-resolution network; and based on the detail refinement network, obtaining a super-resolution image result according to the preliminary up-sampling result. Two networks are cascaded, a task facing a peak signal-to-noise ratio and a task facing subjective feeling are separated and are respectively processed by two models, a basic super-resolution network firstlycompletes preliminary up-sampling, and the result of the preliminary up-sampling ensures content consistency with an original high-resolution image to a great extent; then, a preliminary up-sampling result is sent to a refinement network to further improve sensory loss, and the refinement network has the input of the preliminary up-sampling result, so that the content consistency of the whole super-resolution result is not excessively sacrificed; in addition, the training difficulty of a refinement network and a discriminator is reduced through refinement on a preliminary up-sampling result, the training speed is increased, and the problem of an existing model is effectively relieved.

Description

technical field [0001] The invention belongs to the technical field of image super-resolution, and in particular relates to an image super-resolution method and a cascade network based on a cascade network framework. Background technique [0002] Super-resolution is a technique for reconstructing higher-resolution images or sequences from observed low-resolution images. Due to the diffraction characteristics of light, the photosensitive elements in the imaging system are often unable to obtain details lower than the wavelength of visible light, which limits the detail information that a single pixel can contain. On the whole, a visible light imaging system can be regarded as a system with a certain The low-pass filter of the frequency threshold, all image information above this threshold cannot be effectively recorded by the imaging system. The super-resolution task is to use the complementary ability and correlation between the low-frequency information and high-frequency i...

AI Technical Summary

Problems solved by technology

In existing methods, these weights are hyperparameters that require manual parameter adjustment by humans. Manual parameter adjustment brings about longer model training time and instability in model reconstruction performance.

[0005] (2) The existing methods of improving the perception effect through the transformation of the loss function often bring about the improvement of the perception effect while reducing the performance on the objective indicators, and often generate some perception effects that are good but the peak signal-to-noise ratio. not high image

Images