Gabor is Enough: Interpretable Deep Denoising with a Gabor Synthesis Dictionary Prior

@article{Janjusevic2022GaborIE,
  title={Gabor is Enough: Interpretable Deep Denoising with a Gabor Synthesis Dictionary Prior},
  author={Nikola Janjusevic and Amirhossein Khalilian-Gourtani and Yao Wang},
  journal={2022 IEEE 14th Image, Video, and Multidimensional Signal Processing Workshop (IVMSP)},
  year={2022},
  pages={1-5}
}
Image processing neural networks, natural and artificial, have a long history with orientation-selectivity, often described mathematically as Gabor filters. Gabor-like filters have been observed in the early layers of CNN classifiers and even throughout low-level image processing networks. In this work, we take this observation to the extreme and explicitly constrain the filters of a natural-image denoising CNN to be learned 2D real Gabor filters. Surprisingly, we find that the proposed network… 

Figures and Tables from this paper

CDLNet: Noise-Adaptive Convolutional Dictionary Learning Network for Blind Denoising and Demosaicing

This work proposes an unrolled convolutional dictionary learning network (CDLNet) and demonstrates its competitive denoising and joint denoises and demosaicing performance both in low and high parameter count regimes, and shows that such performance extends to the JDD task and unsupervised learning.

References

SHOWING 1-10 OF 33 REFERENCES

Gabor Convolutional Networks

A new deep model is proposed, termed Gabor Convolutional Networks (GCNs or Gabor CNNs), which incorporates Gabor filters into DCNNs to enhance the resistance of deep learned features to the orientation and scale changes.

Deep Convolutional Dictionary Learning for Image Denoising

A novel framework of deep convolutional dictionary learning (DCDicL), which follows the representation model of DicL strictly, learns the priors for both representation coefficients and the dictionaries, and can adaptively adjust the dictionary for each input image based on its content.

Robust and interpretable blind image denoising via bias-free convolutional neural networks

It is shown that deep convolutional networks systematically overfit the noise levels for which they are trained: when deployed at noise levels outside the training range, performance degrades dramatically; in contrast, a bias-free architecture is shown, obtained by removing the constant terms in every layer of the network, including those used for batch normalization.

CDLNet: Noise-Adaptive Convolutional Dictionary Learning Network for Blind Denoising and Demosaicing

This work proposes an unrolled convolutional dictionary learning network (CDLNet) and demonstrates its competitive denoising and joint denoises and demosaicing performance both in low and high parameter count regimes, and shows that such performance extends to the JDD task and unsupervised learning.

Gabor filter assisted energy efficient fast learning Convolutional Neural Networks

This work reduces the training complexity of CNNs by replacing certain weight kernels of a CNN with Gabor filters, which creates a balanced system that gives better training performance in terms of energy and time, compared to the standalone CNN (without any Gabor kernels), in exchange for tolerable accuracy degradation.

FFDNet: Toward a Fast and Flexible Solution for CNN-Based Image Denoising

The proposed FFDNet works on downsampled sub-images, achieving a good trade-off between inference speed and denoising performance, and enjoys several desirable properties, including the ability to handle a wide range of noise levels effectively with a single network.

Training a V1 Like Layer Using Gabor Filters in Convolutional Neural Networks

This work introduces a lateral inhibitory mechanism in Gabor filters in convolutional neural networks to enhance the performance of the neural network and explore a parameter space and search the best suited parameters using cross validation.

Rethinking the CSC Model for Natural Images

A novel feed-forward network that follows an MMSE approximation process to the CSC model, using strided convolutions is suggested, showing the performance of this supervised architecture to be on par with state of the art methods while using much fewer parameters.

Deep K-SVD Denoising

This work proposes an end-to-end deep architecture with the exact K-SVD computational path, and shows how to overcome difficulties arising in turning the K- SVD scheme into a differentiable, and thus learnable, machine.

Gabor Feature Based Convolutional Neural Network for Object Recognition in Natural Scene

A method of the convolutional neural network combined with Gabor filters for strengthening the learning of texture information called Gabor-CNN is proposed, achieving the recognition rate and significantly outperforming the traditional method based on Bag-of-Words model with SIFT.