Gradient Prior-Aided CNN Denoiser With Separable Convolution-Based Optimization of Feature Dimension

Cho, Sung In; Kang, Suk‐Ju

doi:10.1109/tmm.2018.2859791

Cited by 34 publications

(21 citation statements)

References 26 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In the case of the D, the two fully connected layers are added to the last convolution layer so that the scalar probability value indicating whether the input image is a noise-free image can be derived. For the input of the G and the D, gradients of a given input are used, as shown in [ 19 ]. The gradients are extracted from eight neighborhoods of a current pixel, and there are three kinds of color channels.…”

Section: Methodsmentioning

confidence: 99%

“…The gradients are extracted from eight neighborhoods of a current pixel, and there are three kinds of color channels. Therefore, the total 24 feature channels are used as inputs for the G and the D. In the case of the G, this gradient input can increase its denoising performance [ 19 ]. In the case of the D, this can help the D estimate the fidelity of structural information between the restored and the noise-free images.…”

Section: Methodsmentioning

confidence: 99%

“…Since each feature channel to which the D_conv is applied already contains the combined result of the previous F D1 feature maps, the dilated convolution is a very effective approach for enhancing the denoising performance. Compared to the existing networks that increase the size of the dilation in half of the entire layers and then decrease it in the rest layers [ 1 , 10 , 19 ], the proposed method uses the DSDC module to create multiple cycles that repeat the expansion and contraction of the dilation size to prevent the artifacts that may occur as the size of the dilation becomes too large. Also, the symmetric skip connection can increase the efficiency of information transfer between dilated convolutions.…”

Section: Methodsmentioning

confidence: 99%

“…In our method, the number of layers, F D1 , and F D2 were set to 13, 96, and 96, respectively, by considering the total number of weights for convolution and the quality of image denoising. Specifically, we determined the numbers of layers and F D1 , F D2 values so that only the smaller number of parameters than existing CNNs can be used while providing a comparable to or better denoising quality than the existing CNNs [ 1 , 10 , 19 ]. Even if the number of convolution layers, F D1 , and F D2, are increased, the computational complexity of the proposed method is much smaller than that of the conventional CNN denoisers [ 1 , 10 ] which use the general convolution process because the decrease in computational complexity by the DSC overwhelms the increase in the computational complexity by the increase in the number of layer, F D1 , and F D2 .…”

Section: Methodsmentioning

confidence: 99%

See 3 more Smart Citations

A Generative Adversarial Network-Based Image Denoiser Controlling Heterogeneous Losses

Cho

Park

Kang

2021

Sensors

Self Cite

View full text Add to dashboard Cite

We propose a novel generative adversarial network (GAN)-based image denoising method that utilizes heterogeneous losses. In order to improve the restoration quality of the structural information of the generator, the heterogeneous losses, including the structural loss in addition to the conventional mean squared error (MSE)-based loss, are used to train the generator. To maximize the improvements brought on by the heterogeneous losses, the strength of the structural loss is adaptively adjusted by the discriminator for each input patch. In addition, a depth wise separable convolution-based module that utilizes the dilated convolution and symmetric skip connection is used for the proposed GAN so as to reduce the computational complexity while providing improved denoising quality compared to the convolutional neural network (CNN) denoiser. The experiments showed that the proposed method improved visual information fidelity and feature similarity index values by up to 0.027 and 0.008, respectively, compared to the existing CNN denoiser.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Methodsmentioning

confidence: 99%

Section: Methodsmentioning

confidence: 99%

Section: Methodsmentioning

confidence: 99%

See 2 more Smart Citations

A Generative Adversarial Network-Based Image Denoiser Controlling Heterogeneous Losses

Cho

Park

Kang

2021

Sensors

Self Cite

View full text Add to dashboard Cite

show abstract

“…It can transform one-dimensional (1D) EEG signals into two-dimensional (2D) signals on the complex plane, which is more conducive to capturing the dynamic correlation information of EEG signals. Cho et al [8] proposed an image denoising method based on CNN, which not only improves the denoising performance, but also uses the separable convolution and the gradient prior in this study to reduce the computational complexity. Compared with the existing CNN denoising methods, this method has better denoising quality and is suitable for a variety of image processing works including EEG.…”

Section: Introductionmentioning

confidence: 99%

S-EEGNet: Electroencephalogram Signal Classification Based on a Separable Convolution Neural Network With Bilinear Interpolation

Huang

Xue

et al. 2020

IEEE Access

View full text Add to dashboard Cite

As one of the most important research fields in the brain-computer interface (BCI) field, electroencephalogram (EEG) classification has a wide range of application values. However, for the EEG signal, it is difficult for the traditional neural networks to capture the characteristics of the EEG signal more comprehensively from the time and space dimensions, which has a certain effect on the accuracy of EEG classification. To solve this problem, we can improve the accuracy of classification via end-to-end learning of the time and space dimensions of EEG. In this paper, a new type of EEG classification network, the separable EEGNet (S-EEGNet), is proposed based on Hilbert-Huang transform (HHT) and a separable convolutional neural network (CNN) with bilinear interpolation. The EEG signal is transformed into time-frequency representation by HHT, which allows the EEG signal to be better described in the frequency domain. Then, the depthwise and pointwise elements of the network are combined to extract the feature map. The displacement variable is added by the bilinear interpolation method to the convolution layer of the separable CNN, allowing the free deformation of the sampling grid. The deformation depends on the local, dense, and adaptive input characteristics of the EEG data. The network can learn from the time and space dimensions of EEG signals end to end to extract features to improve the accuracy of EEG classification. To show the effectiveness of S-EEGNet, the team used this method to test two different types of EEG public datasets (motor imagery classification and emotion classification). The accuracy of motor imagery classification is 77.9%, and the accuracy of emotion classification is 89.91%, and 88.31%, respectively. The experimental results showed that the classification accuracy of S-EEGNet improved by 3.6%, 1.15%, and 1.33%, respectively.

show abstract

Ultra-DenseNet for Low-Dose X-Ray Image Denoising in Cardiac Catheter-Based Procedures

Luo

Tóth

Jiang

et al. 2020

Statistical Atlases and Computational Models of the Heart. Multi-Sequence CMR Segmentation, CRT-EPiggy and LV Full Quantificati

View full text Add to dashboard Cite

Gradient Prior-Aided CNN Denoiser With Separable Convolution-Based Optimization of Feature Dimension

Cited by 34 publications

References 26 publications

A Generative Adversarial Network-Based Image Denoiser Controlling Heterogeneous Losses

A Generative Adversarial Network-Based Image Denoiser Controlling Heterogeneous Losses

S-EEGNet: Electroencephalogram Signal Classification Based on a Separable Convolution Neural Network With Bilinear Interpolation

Ultra-DenseNet for Low-Dose X-Ray Image Denoising in Cardiac Catheter-Based Procedures

Contact Info

Product

Resources

About