Method and device for constructing ultra-resolution model of image

Abstract

The invention discloses a method and a device for constructing an ultra-resolution model of an image. The method comprises the following steps of acquiring original images with different resolutions;calculating the conversion coefficient of each original image; performing sampling processing on the corresponding original image according to the conversion coefficient, and obtaining a low-resolution image sample set; inputting the images in the low-resolution image sample set into a three-dimensional convolutional neural network, training the three-dimensional convolutional neural network, andobtaining an ultra-resolution model of the image. Therefore, through the low-resolution image sample set which is generated from the original images with different resolutions, the three-dimensional convolutional neural network is trained for obtaining the ultra-resolution model of the image. Therefore a purpose of reconstructing the high-resolution image through a single model is realized. Furthermore the designed three-dimensional convolutional neural network comprises a plurality of convolutional layers which do not comprise a pooling layer, thereby improving detail distinguishing capability of the reconstructed high-resolution image.

Description

technical field [0001] The present invention relates to the image field, in particular to a method and device for constructing an image super-resolution model. Background technique [0002] The image super-resolution model is a model used to reconstruct a super-resolution image, wherein the super-resolution image reconstruction is to use one or more low-resolution images to obtain a high-resolution image. High-resolution images have higher pixel density and can provide more image details. In the medical field, they have good value for doctors' diagnosis. [0003] The construction of traditional super-resolution images mainly includes three types of methods: interpolation-based methods, reconstruction-based methods, and learning-based methods. The interpolation-based method restores high-resolution images through adjacent pixel information, but does not consider additional high-frequency information; the reconstruction-based method uses prior knowledge of high-resolution tar...

AI Technical Summary

Problems solved by technology

The interpolation-based method restores high-resolution images through adjacent pixel information, but does not consider additional high-frequency information; the reconstruction-based method uses prior knowledge of high-resolution targets to build a model, but the calculation process is complicated; learning-based The method has better adaptability and is currently a better reconstruction method

[0004] However, due to the different resolutions of images obtained by different imaging devices, the current learning-based methods need to use different image super-resolution models for different resolution images to reconstruct high-resolution images, or in other words, cannot achieve One model reconstructs high-resolution images for different resolutions

Images