Image super-resolution reconstruction method based on sparse autoencoder network and extremely fast learning

Abstract

The invention discloses an image super-resolution reconstruction method based on a sparse self-encoding network and extremely fast learning, which solves the problems of inaccurate feature extraction and inaccurate recovery caused by separate feature extraction and reconstruction steps. The main steps are: input a low-resolution test image, block the input low-resolution test image by overlapping, and construct a low-resolution test image block; for each low-resolution image block, pass through a hierarchical sparse autoencoder Extract sparse features; project the sparse features to the high-resolution pixel space through the pre-learned extremely fast learning machine to obtain high-resolution image blocks; finally, aggregate the high-resolution image blocks to obtain the final high-resolution image to complete image super-resolution reconstruction. The invention adopts a multi-layer sparse self-encoding network to learn sparse features, integrates feature extraction and reconstruction into a unified network framework, effectively improves the quality of reconstructed images and improves detail information, and is suitable for super-resolution reconstruction of various natural images.

Description

technical field [0001] The invention belongs to the technical field of image processing, and mainly relates to image super-resolution reconstruction, in particular to an image super-resolution reconstruction method based on sparse self-encoding network and extremely fast learning, which can be used for super-resolution reconstruction of various natural images and character images . Background technique [0002] Image super-resolution reconstruction technology can restore high-resolution images from low-resolution images by means of signal processing, and has been widely used in many fields. Traditional reconstruction-based super-resolution methods mostly use prior knowledge construction constraints such as image edge characteristics, pixel non-negativity, and local smoothness characteristics, and the prior information of the image itself is insufficiently utilized. The resulting image recovered when it is large is too smooth. Freeman et al. proposed a learning-based method...

AI Technical Summary

Problems solved by technology

However, the feature extraction of these methods is not accurate enough, and accurate neighbors are often not found, and the feature extraction and reconstruction steps are performed independently, resulting in inaccurate reconstruction, and the quality of the reconstructed image is average.

Images