Image denoising method based on pixel-level global noise estimation codec network

Abstract

The invention relates to an image denoising method based on a pixel-level global noise estimation encoding and decoding network, which includes: inputting original noisy pictures into the input module of the encoding network, using convolution to perform preliminary feature extraction on the original noisy pictures, and outputting the original features. Figure; process the original feature map through several cascaded coding modules in the coding network, and output the denoised high-level feature map with smaller spatial size and higher semantic level; process the high-level feature map with the The decoding modules in the symmetrical decoding network of the encoding network are processed to obtain a noise-removed output feature map that takes into account both spatial information and semantic information; the output feature map of the decoding network is passed through the output module of the decoding network, Use convolution processing to map the feature dimensions of the output to output the final clear image. This method fully considers the real image noise and global information as well as the characteristics related to its own pixel value, taking into account the denoising effect and running speed.

Description

Technical field [0001] The present invention relates to the technical field of computer vision images, in particular to an image denoising method based on a pixel-level global noise estimation coding and decoding network. Background technique [0002] As an important information carrier, digital images play a more and more important role in daily production and life. Image information is widely used in aerospace, industrial production, military, medical, communications, etc. because of its intuitive and vivid characteristics, including a large amount of information. The field is closely related to our work and life. However, digital images in reality are often affected by the interference of imaging equipment and external environmental noise during the digitization and transmission process, resulting in a decline in image quality and difficulties for subsequent research and processing. Therefore, image denoising, as a basic and important low-level computer vision task, has high ...

AI Technical Summary

Problems solved by technology

However, the above-mentioned deep learning method is aimed at artificially adding Gaussian noise to the synthetic image. The Gaussian noise in the image has nothing to do with the pixel value. It fails to take into account that the noise of the real-world image is not only Gaussian, but also related to the pixel value itself. Correlation, so the above method has poor generalization ability on real noisy images, thus achieving more limited effect

Images