Transform-based twin network image denoising method and system, medium and equipment

Abstract

The invention discloses a twin network image denoising method and system based on Transform, a medium and equipment, and designs two twin networks to extract complementary features, so that the robustness of an obtained denoising device is stronger. Transform is applied to a twin network, saliency features are extracted, a foreground and a background are separated, noise is removed, and a clean image is predicted; a cross interaction mechanism is designed to improve the memory ability of the deep network, and the denoising performance is improved; according to the method, batch normalization, layer normalization, instance normalization, a Swsh function and a linear rectification function activation function component are used in the twin network, so that the learning ability of the denoising network is improved, diversified features can be extracted, the denoising effect is enhanced, and the denoising efficiency is improved. In addition, denoising is carried out only through a 12-layer network, the calculation cost of the network is greatly reduced, and the requirements of mobile equipment are met very well. And saliency features can be adaptively extracted according to different scenes, and the method has a blind denoising function and a relatively high practical application value.

Description

technical field [0001] The invention belongs to the technical fields of deep learning, image restoration and computer vision, and in particular relates to a Transformer-based twin network image denoising method, system, medium and equipment. Background technique [0002] With the rapid development of today's information age, digital image equipment has been widely used in many fields such as UAV disaster rescue, face recognition, ocean detection, etc., but it is often affected by vibration, noise, shooting environment, etc. The collected images are noisy due to other interferences. Therefore, image denoising technology has important practical application value. [0003] Convolutional Neural Network (CNN) has been widely used in image denoising due to its super learning ability, but most CNNs only rely on end-to-end network structure to extract effective features and complete image denoising tasks. It will increase the complexity of training the network. Contents of the i...

AI Technical Summary

Problems solved by technology

[0004] The technical problem to be solved by the present invention is to provide a Transformer-based twin network image denoising method and system for the above-mentioned deficiencies in the prior art, extract complementary information, make the extracted features more expressive, and integrate Transformer into In the Siamese network, salient features are extracted, noise information is extracted from complex backgrounds, and image denoising tasks are efficiently completed, which is suitable for solving image denoising problems in complex scenes

Images