Single image denoising method and device based on cavitated kernel prediction network

Abstract

The invention discloses a single image denoising method and device based on cavitated kernel prediction network. The method comprises the following steps: constructing the cavitated kernel prediction network comprising a feature extraction module, a feature compression module, a kernel prediction module and an image reconstruction module; performing parameter optimization on the cavitated kernel prediction network for later use; during application, inputting a noise image is input into the cavitated kernel prediction network after parameter optimization, extracting an advanced feature map from the noise image by using the feature extraction unit, compressing the advanced feature map into a compressed feature map by using the feature compression module, extracting a prediction convolution kernel according to the advanced feature map by using the kernel prediction module, obtaining a prediction image based on the prediction convolution kernel and the compressed feature image by using an image reconstruction module, and fusing the prediction image with the noise image to obtain a denoised image. A kernel prediction network is introduced for a single image denoising problem, and techniques such as cavity convolution, multi-kernel channel fusion and feature map compression are used to realize a single image denoising task, so that the denoising effect is greatly improved.

Description

technical field [0001] The invention relates to the field of computer science image processing, in particular to a method and device for denoising a single image based on a hollowed-out kernel prediction network. Background technique [0002] Image denoising is a fundamental problem in the field of image processing. In recent years, the rapid development of deep learning networks provides an efficient solution for denoising algorithms. However, the traditional convolutional network uses the same convolution kernel on the pixels of the entire image, such as the joint noise estimation and image denoising method based on deep learning disclosed in the application publication number CN109658348A, which does not make full use of the feature differences between pixel regions. [0003] Although there have been some newer studies trying to introduce attention mechanism to emphasize this local difference, such as an image adaptive denoising method based on attention mechanism disclos...

AI Technical Summary

Problems solved by technology

However, this method has not been well applied to the single image denoising problem, and its main difficulty lies in extracting enough information from the only reference image for pixel-by-pixel kernel prediction

Images