Unsupervised image denoising method based on uncertainty perception

Abstract

The invention belongs to the technical field of digital image processing, and particularly relates to an unsupervised image denoising method based on uncertainty perception. The method comprises the following steps: constructing a smooth noise extraction module and a rough denoising network to realize simulation of a noise block and rough denoising of an image; a network with modeling accidental uncertainty is constructed, pixels with high uncertainty in the noise image are estimated, an uncertainty map is generated, and the network in the next stage is guided to carry out denoising; and constructing a fine denoising network, and guiding the network to reduce the influence of uncertainty on the denoising process through an uncertainty graph weighting objective function, so that the network achieves a better denoising effect. Experimental results show that noise in the image can be effectively removed, and the denoised image has a good texture structure and visual quality.

Description

technical field [0001] The invention belongs to the technical field of digital image processing, and in particular relates to an image denoising method. Background technique [0002] Image denoising, as a classic and challenging image processing task, has received more and more attention. The noise in the image is mainly generated in the process of acquiring the image. Two common types of image sensors, CCD and CMOS, will introduce various noises, such as thermal noise caused by resistors, and channel heat of FETs, due to their own sensor material properties, working environment, electronic components, and circuit structures. Noise, photon noise, dark current noise, photoresponse non-uniformity noise. The presence of noise will not only degrade the visual quality of the picture, but even bring about more serious effects. For example, low-dose radiation CT images have noise due to the reduced radiation intensity, and these noises will affect the doctor's judgment on the le...

AI Technical Summary

Problems solved by technology

Although the above methods have good performance on denoising tasks, these methods tend to generate the mean of multiple possible denoising results, and do not consider the impact of accidental uncertainty in the training data set on the denoising network, resulting in network The denoising result is too smooth and the visual effect is poor

Images