http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
Residual Learning of Transfer-learned AlexNet for Image Denoising
Mohan Laavanya,Veeramani Vijayaraghavan 대한전자공학회 2020 IEIE Transactions on Smart Processing & Computing Vol.9 No.2
In today’s scenarios, deep learning has fascinated all researchers from numerous arenas who developed ways to achieve obligatory outcomes. In deep learning, transfer learning is undergoing deep study, because the study helps to practice a pre-trained network for our own tasks. A novel, transfer-learned AlexNet-based residual learning for Gaussian noise reduction is presented in this paper. The method can remove any level of Gaussian noise without having information about the noise variance in both gray scale and color images. Therefore, our technique is blind Gaussian image denoising that learns a residual image by eradicating the clean image from the transfer-learned AlexNet, and removes noise by identifying the difference from the input image. Experimental results with the proposed scheme are compared against a Gaussian denoiser for image denoising in terms of peak signal-to-noise ratio (PSNR) and visual perception. The results have revealed that our residual learning using transfer-learned AlexNet attains promising denoising results.
Image Denoising with a Convolution Neural Network using Gaussian Filtered Residuals
Laavanya Mohan,Vijayaraghavan Veeramani 대한전자공학회 2021 IEIE Transactions on Smart Processing & Computing Vol.10 No.2
Deep learning using a convolutional neural network has become a state-of-art technique in image processing. In recent scenarios, image denoising using a residual image in deep learning has been popular. However, one aspect missing in these methods is that the residual image has all the noise and very small structured details of the input image. Therefore, we have developed a Gaussian filter residual convolutional neural network architecture for color image denoising. Gaussian residual learning was used to boost the denoising performance. The architecture is designed to remove additive white Gaussian noise, which is one of the most basic types of noise that affects an image when captured. The network with Gaussian residual learning removes the clean image using the features learned from the hidden layer. The peak signal-to-noise ratio and structural similarity index measure achieved by our method reveals that the presented approach is better at denoising images with Gaussian noise than a convolutional neural network.