Abstract
In recent years, convolutional neural networks have achieved considerable success in different computer vision tasks, including image denoising. In this work, we present a residual dense neural network (RDUNet) for image denoising based on the densely connected hierarchical network. The encoding and decoding layers of the RDUNet consist of densely connected convolutional layers to reuse the feature maps and local residual learning to avoid the vanishing gradient problem and speed up the learning process. Moreover, global residual learning is adopted such that, instead of directly predicting the denoised image, the model predicts the residual noise of the corrupted image. The algorithm was trained for the case of additive white Gaussian noise and using a wide range of noise levels. Hence, one advantage of the proposal is that the denoising process does not require prior knowledge about the noise level. In order to evaluate the model, we conducted several experiments with natural image databases available online, achieving competitive results compared with state-of-the-art networks for image denoising. For comparison purpose, we use additive Gaussian noise with levels 10, 30, 50. In the case of grayscale images, we achieved PSNR of 34.39, 29.11, 26.99, and SSIM of 0.9297, 0.8193, 0.7491. For color images we obtained PSNR of 36.68, 31.43, 29.12, and SSIM of 0.9600, 0.8961, 0.8465.
Highlights
Image denoising is an important problem in the area of lowlevel image processing
We present a residual dense neural network (RDUNet) for image denoising with competitive results with state of the art
We present a convolutional neural network based on Denoising hierarchical image denoising network (DHDN) model for natural image denoising, which is capable of handling a wide range of Gaussian noise levels
Summary
Image denoising is an important problem in the area of lowlevel image processing. The main objective of image denoising problem is to recover a clean image x, which has been corrupted by some noise v from a source. Some recent neural networks for image denoising, with fewer parameters, have been proposed to achieve competitive results [8]–[10]. These models are trained for a specific noise level, requiring a model instance for every noise level. The U-Net model’s architecture consists of a contracting (encoder) path to capture context and a symmetric expanding path (decoder) to estimate the segmentation This architecture has recently been used for image denoising in models such as multilevel wavelet CNN (MWCNN) [12], densely connected. We present a convolutional neural network based on DHDN model for natural image denoising, which is capable of handling a wide range of Gaussian noise levels.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have