A5011844226- Image and Signal Denoising Methods
- Advanced Image Processing Techniques
- Medical Image Segmentation Techniques
- Image Processing Techniques and Applications
- Advanced Neural Network Applications
- Radiomics and Machine Learning in Medical Imaging
- Sparse and Compressive Sensing Techniques
- Tensor decomposition and applications
- Domain Adaptation and Few-Shot Learning
- Advanced Vision and Imaging
- Crystallization and Solubility Studies
- Medical Imaging and Analysis
- AI in cancer detection
- Atmospheric Ozone and Climate
- Topic Modeling
- Advanced Control Systems Design
- Spectroscopy Techniques in Biomedical and Chemical Research
- Advanced Control Systems Optimization
- Statistical Methods and Inference
- NMR spectroscopy and applications
- Astrophysics and Cosmic Phenomena
- Photocathodes and Microchannel Plates
- Spaceflight effects on biology
- Advanced Neuroimaging Techniques and Applications
- Generative Adversarial Networks and Image Synthesis
University of Oxford
2024
Fudan University
2019-2023
Wuhan University
2017-2020
Chinese Academy of Sciences
2012
National Space Science Center
2010-2012
This paper reviews the NTIRE 2021 challenge on learning super-Resolution space. It focuses participating methods and final results. The addresses problem of a model capable predicting space plausible super-resolution (SR) images, from single low-resolution image. must thus be sampling diverse outputs, rather than just generating SR goal is to spur research into developing formulations models better suited for highly ill-posed problem. And thereby advance state-of-the-art in broader field. In...
Image decomposition aims to analyze an image into elementary components, which is essential for numerous downstream tasks and also by nature provides certain interpretability the analysis. Deep learning can be powerful such tasks, but surprisingly their combination with a focus on generalizability rarely explored. In this work, we introduce novel framework interpretable deep decomposition, combining hierarchical Bayesian modeling create architecture-modularized model-generalizable neural...
Single image super-resolution (SR) is extremely difficult if the upscaling factors of pairs are unknown and different from each other, which common in real SR. To tackle difficulty, we develop two multi-scale deep neural networks (MsDNN) this work. Firstly, due to high computation complexity high-resolution spaces, process an input mainly downscaling could greatly lower usage GPU memory. Then, reconstruct details image, design a residual network (MsRN) spaces based on blocks. Besides,...
Modeling statistics of image priors is useful for super-resolution, but little attention has been paid from the massive works deep learning-based methods. In this work, we propose a Bayesian restoration framework, where natural are modeled with combination smoothness and sparsity priors. Concretely, firstly consider an ideal as sum component residual, model real degradation including blurring, downscaling, noise corruption. Then, develop variational approach to infer their posteriors....
Segmentation is a critical step in analyzing the developing human fetal brain.There have been vast improvements automatic segmentation methods past several years, and Fetal Brain Tissue Annotation (FeTA) Challenge 2021 helped to establish an excellent standard of brain segmentation.However, FeTA was single center study, generalizability algorithms across different imaging centers remains unsolved, limiting real-world clinical applicability.The multi-center 2022 focuses on advancing for...
Due to the cross-domain distribution shift aroused from diverse medical imaging systems, many deep learning segmentation methods fail perform well on unseen data, which limits their real-world applicability. Recent works have shown benefits of extracting domain-invariant representations domain generalization. However, interpretability features remains a great challenge. To address this problem, we propose an interpretable Bayesian framework (BayeSeg) through modeling image and label...
The principal rank-one (RO) components of an image represent the self-similarity image, which is important property for restoration. However, RO a corrupted could be decimated by procedure denoising. We suggest that should utilized and decimation avoided in To achieve this, we propose new framework comprised two modules, i.e., decomposition reconstruction. developed to decompose into residual. This achieved successively applying projections or its residuals extract components. projections,...
An imaging system based on single photon counting and compressive sensing (ISSPCCS) is developed to reconstruct a sparse image in absolute darkness. The avalanche detector spatial light modulator (SLM) of aluminum micro-mirrors are employed the while convex optimization used reconstruction algorithm. an object very dark can be reconstructed from under-sampling data set, but with high SNR robustness. Compared traditional single-pixel camera photomultiplier tube (PMT) as detector, ISSPCCS...
Super-resolution (SR) is an ill-posed problem, which means that infinitely many high-resolution (HR) images can be degraded to the same low-resolution (LR) image. To study one-to-many stochastic SR mapping, we implicitly represent non-local self-similarity of natural and develop a Variational Sparse framework for Super-Resolution (VSpSR) via neural networks. Since every small patch HR image well approximated by sparse representation atoms in over-complete dictionary, design two-branch...
Contrast-enhanced brain MRI (CE-MRI) is a valuable diagnostic technique but may pose health risks and incur high costs. To create safer alternatives, multi-modality medical image translation aims to synthesize CE-MRI images from other available modalities. Although existing methods can generate promising predictions, they still face two challenges, i.e., exhibiting over-confidence lacking interpretability on predictions. address the above this paper introduces TrustI2I, novel trustworthy...
Single image super-resolution (SR) is extremely difficult if the upscaling factors of pairs are unknown and different from each other, which common in real SR. To tackle difficulty, we develop two multi-scale deep neural networks (MsDNN) this work. Firstly, due to high computation complexity high-resolution spaces, process an input mainly downscaling could greatly lower usage GPU memory. Then, reconstruct details image, design a residual network (MsRN) spaces based on blocks. Besides,...
Registration networks have shown great application potentials in medical image analysis. However, supervised training methods a demand for large and high-quality labeled datasets, which is time-consuming sometimes impractical due to data sharing issues. Unsupervised registration algorithms commonly employ intensity-based similarity measures as loss functions without any manual annotations. These estimate the parameterized transformations between pairs of moving fixed images through...
Deep space life exploration nowadays mainly focuses on whether environments other planets are suitable for the existence of life. This has not hit key point. As depending its specified conditions, it is great necessity and importance to conduct in life's original environment. An idea was put forward this paper detect specific forms environment if they exist. idea, based fluorescence, included light source unit detection unit. The former optical fiber coupled LED lens; latter charge device...
Cherenkov radiation is used to study the production of particles during collisions, cosmic rays detections and distinguishing between different types neutrinos electrons. The optical properties water are very important research Effect. Lambert-beer law a method attenuation light through medium. In this paper, investigated by use performance test system. system composed light-emitting diode (LED) source photon receiver models. LED model provides pulse signal which frequency 1 kHz width 100ns....
Medical images are generally acquired with limited field-of-view (FOV), which could lead to incomplete regions of interest (ROI), and thus impose a great challenge on medical image analysis. This is particularly evident for the learning-based multi-target landmark detection, where algorithms be misleading learn primarily variation background due varying FOV, failing detection targets. Based learning navigation policy, instead predicting targets directly, reinforcement (RL)-based methods have...
The charged colloidal crystal is one of the hotspots on condensed physics research. main reason forming that particles achieve a balance and long orderly arrangement because static repelling force between granular Brownian movement. However, slight fluctuations temperature will cause strength motion changes then influence crystallization state process. This paper solves key work research phase dynamics: solution high precision low volatility control system. A proportional-integral-derivative...