Research On Denoising Technology Of Optical Encrypted Image Based On Deep Learning

Image denoising has always been a hot issue in image processing tasks.The image signal is generated due to the interference from the receiving signal equipment and external factors during the transmission process.Image noise affects people's subjective visual perception.For task areas with high image quality standards such as medical images and satellite images,the requirements for denoising are strict.Existing denoising algorithms can remove low-intensity noise signals well,but cannot effectively remove high-intensity noise.The optical image encryption system in Joint Optical Conversion Technology(JTC)has a simple structure and does not require the production of a complex phase difference key.The encrypted image is an intensity image,which is appropriately recorded and transmitted.However,the main disadvantage of this encryption method is The quality of the decrypted image is poor and there is severe noise.Traditional denoising algorithms can eliminate noise to a certain extent.Aiming at the problem of noise interference in optical encryption systems and poor denoising effects of traditional algorithms,deep learning technology is used here to remove noise interference in optical encrypted images.For the optically encrypted noise image,a new feature extraction module is proposed here.This module uses dense modules to enhance the replacement of feature information,enhances the performance of the network,and adds a channel attraction mechanism to extract the overall The network model uses the U-Net network structure to ensure that the low-level information in the image is retained,and a large amount of information and local information are merged to restore high-quality images.At the same time,the idea of confronting the network is dated.The generation network part adopts the above-mentioned model structure,and the multi-scale discriminator is added to the discrimination network to enhance the ability of the network to extract features and restore the reduced noise-free image.The loss function part uses a modified combination to reduce the difference between the original image and the noise,and combines the feature extraction network VGG-19 network to extract the detailed features of the image.The comparison between the subjective observation of the human eye and theevaluation standard of objective image proves the effectiveness of this method.