A5047634961- Image and Signal Denoising Methods
- Advanced Image Processing Techniques
- Sparse and Compressive Sensing Techniques
- Seismic Imaging and Inversion Techniques
- Photoacoustic and Ultrasonic Imaging
- Image and Video Quality Assessment
- Advanced Vision and Imaging
- Underwater Acoustics Research
- Advanced Image Fusion Techniques
- Medical Imaging Techniques and Applications
- Advanced X-ray and CT Imaging
- Geophysical Methods and Applications
Ulsan National Institute of Science and Technology
2018-2020
York University
2019
This paper reviews the NTIRE 2019 challenge on real image denoising with focus proposed methods and their results. The has two tracks for quantitatively evaluating performance in (1) Bayer-pattern raw-RGB (2) standard RGB (sRGB) color spaces. had 216 220 registered participants, respectively. A total of 15 teams, proposing 17 methods, competed final phase challenge. by teams represent current state-of-the-art targeting noisy images.
Recently developed deep-learning-based denoisers often outperform state-of-the-art conventional such as the BM3D. They are typically trained to minimize mean squared error (MSE) between output image of a deep neural network (DNN) and ground truth image. Thus, it is important for use high quality noiseless data performance. However, challenging or even infeasible obtain images in some applications. Here, we propose method based on Stein's unbiased risk estimator (SURE) training DNN only noisy...
Compressive sensing is a method to recover the original image from undersampled measurements. In order overcome ill-posedness of this inverse problem, priors are used such as sparsity, minimal total-variation, or self-similarity images. Recently, deep learning based compressive recovery methods have been proposed and yielded state-of-the-art performances. They data-driven approaches instead hand-crafted regularize ill-posed problems with data. Ironically, training neural networks (DNNs) for...
This paper reviews the NTIRE 2022 Challenge on Super-Resolution and Quality Enhancement of Compressed Video. In this challenge, we proposed LDV 2.0 dataset, which includes dataset (240 videos) 95 additional videos. challenge three tracks. Track 1 aims at enhancing videos compressed by HEVC a fixed QP. 2 3 target both super-resolution quality enhancement video. They require x2 x4 super-resolution, respectively. The tracks totally attract more than 600 registrations. test phase, 8 teams, teams...
Recently, deep neural network (DNN) based methods for low-dose CT have been investigated to achieve excellent performance in both image quality and computational speed. However, almost all using DNNs require clean ground truth data with full radiation dose train the DNNs. In this work, we attempt reconstructions reduced tube current by investigating unsupervised training of denoising sensor measurements or sinograms without full-dose images. other words, our proposed allow only noisy...
Recently, Stein's unbiased risk estimator (SURE) has been applied to unsupervised training of deep neural network Gaussian denoisers that outperformed classical non-deep learning based and yielded comparable performance those trained with ground truth. While SURE requires only one noise realization per image for training, it does not take advantage having multiple realizations when they are available (e.g., two uncorrelated Noise2Noise). Here, we propose an extended (eSURE) train correlated...
Recently, there have been several works on unsupervised learning for training deep based denoisers without clean images. Approaches Stein's unbiased risk estimator (SURE) shown promising results Gaussian denoisers. However, their performance is sensitive to hyper-parameter selection in approximating the divergence term SURE expression. In this work, we briefly study computational efficiency of Monte-Carlo (MC) approximation over recently available exact computation using backpropagation....
Compressive sensing is a method to recover the original image from undersampled measurements. In order overcome ill-posedness of this inverse problem, priors are used such as sparsity in wavelet domain, minimum total-variation, or self-similarity. Recently, deep learning based compressive recovery methods have been proposed and yielded state-of-the-art performances. They data-driven approaches instead hand-crafted solve ill-posed problem with data. Ironically, training neural networks for...
This paper reviews the NTIRE 2022 Challenge on Super-Resolution and Quality Enhancement of Compressed Video. In this challenge, we proposed LDV 2.0 dataset, which includes dataset (240 videos) 95 additional videos. challenge three tracks. Track 1 aims at enhancing videos compressed by HEVC a fixed QP. 2 3 target both super-resolution quality enhancement video. They require x2 x4 super-resolution, respectively. The tracks totally attract more than 600 registrations. test phase, 8 teams, teams...