Real image denoising method based on multi-scale selection feedback network

Abstract

The invention discloses a real image denoising method based on a multi-scale selection feedback network. The method comprises the following steps: constructing a multi-scale selection module MSB for extracting multiple receptive field scale features; constructing a multi-scale selection feedback network MSFB, wherein the MSFB comprises a shallow feature extraction unit, a plurality of MSBs connected in series, an image reconstruction unit and a degradation model; for image denoising, two dual tasks are constructed: predicting a noise-free image from an original noise image, and degrading the predicted noise-free image to a noise image; repeatedly executing two dual tasks in a plurality of time steps by using the MSFB, and performing multi-stage iteration; in iteration, selectively feeding back high-level semantic information output by the deep MSB of the previous time step to the input end of the shallow MSB of the next time step, and the MSFB is trained through iteration; the training process taking minimization of dual loss as an optimization target and taking a peak signal-to-noise ratio as an evaluation index of network performance; and inputting a noise image into the trained MSFB for de-noising, and outputting the de-noised image.

Description

technical field [0001] The invention relates to the fields of computer vision and image processing, in particular to a real image denoising method based on a multi-scale selective feedback network. Background technique [0002] A variety of complex noises will be generated during the processing, storage, and transmission of real images in the acquisition system, resulting in the loss of structural details and the degradation of image quality. The image noise will also be subject to such decomposition and synthesis. The electrical system and external influences in these processes complicate the precise analysis of image noise. Most existing denoising methods are based on known synthetic additive white Gaussian noise, but often perform poorly in real-world noisy images. [0003] Image denoising is a typical image restoration task. The nature of direct image-to-image conversion results in infinitely many correspondences of noisy images in the clean domain. This ill-posed prob...

AI Technical Summary

Problems solved by technology

[0006] In view of this, the present invention proposes a real image denoising method based on a multi-scale selective feedback network, adding additional supervision in the noise domain to the network, which not only reduces the excessive dependence on clean and high-quality training data, but also makes the network sensitive to noise changes. It is more robust to solve the problems of poor denoising effect and high complexity of the denoising model in current denoising methods for real noisy images

Images