Deep attention coding and decoding single-image super-resolution algorithm based on perception loss guidance

Abstract

The invention provides a deep attention coding and decoding single-image super-resolution algorithm based on perception loss guidance. The deep attention coding and decoding single-image super-resolution algorithm comprises the following steps: constructing a deep attention coding and decoding network model based on perception loss guidance; designing a perception loss objective function accordingto the network model; presetting network model hyper-parameters, and training a network model by adopting a back propagation algorithm according to the perception loss objective function; if the network model converges, outputting an expected high-resolution image, otherwise, returning to execute the previous step until the network model converges. According to the invention, the residual space attention unit is added into the network and is used for capturing and reconstructing more low-frequency information; the perceptual loss composed of an average absolute error and structural similarityloss is used as a loss function to optimize network parameters, so that the network can pay more attention to the visual sensitive area to store the visual information structure, the quality of the reconstructed image is improved, the reconstructed image has an excellent visual effect, and the reconstruction efficiency of the network is extremely high.

Description

technical field [0001] The invention belongs to the technical field of image information processing, and in particular relates to a deep attention encoding and decoding single image super-resolution algorithm guided by perceptual loss. Background technique [0002] Image super-resolution is to convert low-resolution images to high-resolution through certain algorithms. High-resolution images have higher pixel density, more detailed information, and more delicate picture quality. To obtain high-resolution images, the most direct way is to use high-resolution cameras. However, in practical applications, due to the consideration of manufacturing process and engineering costs, high-resolution and super-resolution cameras will not be used in many occasions. A camera is used to collect image signals. [0003] The concept and method of super-resolution technology was first proposed by Harris and Goodman in the 1960s. Subsequently, many people have studied it, and have proposed m...

AI Technical Summary

Problems solved by technology

However, most deep learning methods belong to the supervised learning method. These methods require a large number of image pairs, consisting of LR images and corresponding HR images, to pre-train network parameters, which limits the applicability of these methods in actual scenarios.

In some practical problems, it is difficult to collect real HR images, and there are even cases where they cannot be obtained

Images