Image moire elimination method based on deep multi-resolution network

Abstract

The invention aims to provide an image moire elimination method based on a deep multi-resolution network. The method includes: adjusting the depth multi-resolution model; redesigning a down-sampling module, replacing a down-sampling pooling layer with a convolution layer with a step length of 2, carrying out learnable multi-resolution feature sampling on the moire patterns in the input image through the convolution layer with the step length of 2 by the down-sampling module, and enabling a sampled feature map to contain moire pattern information and position information under different resolutions. The down-sampling mode avoids the situation that a backbone network only eliminates moire in a single frequency domain, and due to the fact that the moire exists in a plurality of frequency domains, the elimination effect is poor due to the fact that moire elimination is conducted under a single resolution.

Description

technical field [0001] The invention belongs to the field of image processing, and relates to an image moiré elimination method, in particular to an image moiré elimination method based on a deep multi-resolution network. Background technique [0002] Due to the limitations of the cost and volume of the mobile phone camera itself, the captured images have varying degrees of distortion, especially on some regular texture images, often see ripple-like stripes, that is, moiré. The sampling frequency of the mobile phone camera is fixed. When the high-frequency components of the scene information are rich and do not satisfy the sampling law, if you want to obtain high-quality and clear imaging, there will inevitably be moiré interference when the mobile phone camera shoots the scene. Moiré often has a large area in the image, and the color shift is obvious, which seriously affects the image quality and image analysis results; The interpolation algorithm is closely related, and t...

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to provide an image moiré elimination method based on a deep multi-resolution network, so as to solve the problem that the current super-resolution method based on a deep convolutional neural network is applied to image moiré elimination.

Images