Deep learning image denoising method integrating multiple scales and attention mechanism

Abstract

The invention discloses a deep learning image denoising method integrating multiple scales and an attention mechanism. A peak signal-to-noise ratio and the structural similarity of Gaussian denoisingoutput of a deep learning model to an image are improved. The method mainly comprises the following steps: selecting an appropriate high-definition image training set, and making a corresponding noiseimage; building a deep learning network model and combining a multi-scale mechanism and an attention mechanism; training by using the selected training set and the built deep learning network model and taking the minimized loss function as a target until the loss function converges; and inputting a to-be-denoised image in the test set into the trained denoising network to obtain a denoised image.Compared with a traditional denoising method and an existing deep learning denoising method, the multi-scale and attention mechanism integrated deep learning denoising scheme provided by the invention has the advantage that the peak signal to noise ratio (PSNR) index is obviously improved.

Description

technical field [0001] This program designs the field of image denoising, especially involving multi-scale and attention mechanisms based on deep learning. Background technique [0002] In reality, digital images are often affected by imaging equipment and external environmental noise interference during digitization and transmission, which are called noisy images or noisy images. Image denoising refers to the process of reducing noise in digital images. This is a very classic inverse problem in image processing. Taking additive Gaussian white noise as an example, many early methods use filtering methods to directly process images, such as mean filtering , Median filtering, Wiener filtering, etc. These filtering methods are relatively simple, making the image details restored after filtering processing poor. [0003] Subsequently, the traditional algorithm of BM3D appeared, which filtered by finding similar blocks in the image. The BM3D algorithm is divided into two steps ...

AI Technical Summary

Problems solved by technology

[0005] Although the image denoising technology based on deep learning (DnCNN, etc.) has a certain improvement in PSNR and SSIM indicators compared with traditional image denoising technologies (BM3D, WNNM, etc.), it is still still difficult for certain texture-rich images. than traditional methods

Moreover, the existing network has the disadvantages of a single network structure and no joint use of multiple mechanisms

Images