An Image Denoising Method Based on Visual Perception
Zixuan Wu, Wei Liu, Weidong Chen · 2023
To address the problem of image quality degradation caused by noise during image acquisition and transmission, this article proposes an improved image denoising model based on Generative Adversarial Networks (GAN). The generator network of this model uses a combination of convolutional blocks, residual blocks, and deconvolutional blocks to efficiently extract low-level and high-level feature information to avoid loss of image details and effectively alleviate the problem of gradient disappearance in the network. To improve discrimination accuracy, the discriminator network is adjusted on the basis of the VGG network to achieve pixel-level classification. The loss function consists of a composite loss function composed of Mean Square Error (MSE) loss, visual perception loss, and adversarial loss, considering both the image denoising ability and the preservation of image detail information. Experimental results show that the proposed method outperforms mainstream image denoising techniques in terms of PSNR, SSIM, and visualization effects, indicating that the model can effectively remove image noise while making the image more in line with human visual perception.