Quick image super-resolution reconstruction method

Abstract

The invention discloses a quick image super-resolution reconstruction method. The method comprises the steps of segmenting an image in a training image database into image blocks by utilizing a priori that natural image textures have similarity, and stretching the image blocks into vectors; then training a sparse code dictionary by utilizing the vectors; training the sparse code dictionary for corresponding high and low-resolution image data sets to obtain corresponding high and low-resolution dictionaries; and for a new input low-resolution image, calculating a sparse code by utilizing the low-resolution dictionary first and then multiplying the high-resolution dictionary by the code to obtain a high-resolution image. In a spares coefficient solving process, sparse coefficients of standard orthogonal basis are calculated for a fixed dictionary first, and for a new input sparse coefficient, the sparse coefficient is quickly solved in a manner of weighted summation of the sparse coefficients of the standard orthogonal basis. In an image block processing process, the image blocks are processed in parallel by adopting a CUDA technology, so that the processing time can be shortened to one ten-thousandth of conventional CPU calculation. The method is used for quickly obtaining a super-resolution image of a single image.

Description

technical field [0001] The invention relates to a fast image super-resolution reconstruction method, which belongs to the technical field of computer vision, and specifically relates to sparse representation theory, fast sparse coding solution technology, CUDA parallel acceleration technology and single image super-resolution method. Background technique [0002] People's pursuit of image quality is endless, and the key factor determining image quality is image resolution. Single image super-resolution is to reconstruct a high-resolution image from a single frame of low-resolution image. Super-resolution images not only increase the size of the image, but also have better identification of structural information and texture details than existing images. The traditional image interpolation algorithm assumes that the image has continuity, which makes the calculated high-resolution image blurred in the high-frequency area. [0003] In order to obtain better high-frequency det...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the time-consuming problem of the Yang method, the present invention proposes a fast image super-resolution reconstruction method

Images