Single-image super-resolution reconstruction algorithm based on optical flow method and sparse neighbor embedding

Abstract

The invention discloses a single-image super-resolution reconstruction algorithm based on an optical flow method and sparse neighbor embedding. The single-image super-resolution reconstruction algorithm comprises the following steps: firstly, training and learning offline dictionaries: learning corresponding characteristics of brightness components of high / low-resolution images in a training image set; extracting oriented gradient histograms and gradient characteristics of image blocks, carrying out gradient characteristic dimension reducing processing and dividing to obtain a plurality of clustering subsets to form trained dictionaries; at a reconstruction phase, carrying out RGB color to YCbCr space conversion on the low-resolution images; amplifying chrominance components as reconstructed chrominance components; carrying out bi-cubic interpolation amplification on the brightness components; extracting image characteristics and sequentially matching with a plurality of neighborhood image blocks; calculating an optical flow velocity vector and weighting and combining the plurality of neighborhood image blocks to obtain a final matched result; deblurring reconstructed images and carrying out back projection iteration processing to obtain brightness components of a final reconstruction result; and converting the reconstructed images from a YCbCr color space to an RGB color space. The single-image super-resolution reconstruction algorithm has the advantages that the matching of the image blocks is relatively accurate and the super-resolution reconstruction is relatively effective.

Description

technical field [0001] The invention belongs to the field of digital image processing, and is a single image super-resolution reconstruction algorithm based on optical flow method and sparse neighborhood embedding. Background technique [0002] With the development of the information age, people have higher and higher requirements for the resolution of digital images. The traditional method of improving resolution by improving the quality of imaging components is limited by factors such as technological level and cost price, and cannot continue to meet people's needs. Therefore, image super-resolution algorithms that obtain high-resolution images by means of information processing have gradually become an important topic in the field of modern image processing. [0003] Generally, the image degradation model can be expressed as: [0004] Y=DHX+n (1) [0005] Among them, D is the downsampling matrix, H is the fuzzy matrix, and n is the noise term. [0006] The current sup...

AI Technical Summary

Problems solved by technology

The performance of these algorithms is different, but without exception, they all regard the dictionary image block as a fixed vector, and the dictionary image block cannot dynamically adjust its geometric structure characteristics during the reconstruction process to adapt to low-resolution input image, so it is impossible to achieve a greater improvement in matching accuracy

Moreover, most of the current algorithms cannot adaptively realize adjacent image blocks that contribute to neighborhood matching, and cannot minimize the matching error.

In addition, the current algorithm is too simple when processing the low-frequency information of the image. Most of them just subtract their own mean value to complete the low-frequency process. This method cannot fully eliminate the interference from the low-frequency basic content in the original image.

Images