Distributed image super-resolution method based on k-means driven convolutional neural network

Abstract

The present invention provides a distributed image super-resolution method based on K-means-driven convolutional neural network, the method comprising: cutting out the low-resolution image to be super-resolved according to the position to obtain a plurality of low-resolution small image blocks; extracting multiple The content structure characteristics of a low-resolution small image block; according to the content structure feature, use the K cluster centers of the training model to classify multiple low-resolution small image blocks; perform up-sampling on multiple low-resolution small image blocks , to obtain a small image block with high resolution and no details; input the small image block with high resolution and no details corresponding to the small image block with low resolution into the convolutional neural network model corresponding to the cluster center to which it belongs, and obtain the output of the model, A high-resolution small image block is obtained by adding the output to a high-resolution small image without details; a plurality of high-resolution small image blocks are spliced ​​according to positions to obtain a final high-resolution image. The distributed image super-resolution method of the present invention has a high super-resolution effect.

Description

technical field [0001] The invention relates to image super-resolution technology, in particular to a distributed image super-resolution method based on K-means driving convolutional neural network. Background technique [0002] Single image super-resolution is a classic topic in the field of computer vision, and its purpose is to restore a high-resolution image with more details from a single low-resolution image. Single image super-resolution is widely used in various applications in the field of computer vision, such as security monitoring imaging, medical imaging and other fields that require more image details. The emergence of single image super-resolution makes up for the defects of low hardware resolution of image acquisition equipment and low resolution of imaging targets caused by long-distance shooting, and forms high-quality images. [0003] In the case that the image acquisition equipment does not capture high-resolution images, using single image super-resolut...

AI Technical Summary

Problems solved by technology

[0004] However, existing single image super-resolution algorithms such as Bicubic, SC, ANR, SRCNN, etc., have not achieved the optimal super-resolution effect due to the limited learning ability of their own algorithms.

Images