A5101848673- Advanced Image Processing Techniques
- Image and Signal Denoising Methods
- Adversarial Robustness in Machine Learning
- Advanced Image Fusion Techniques
- Image Processing Techniques and Applications
- Remote-Sensing Image Classification
- Anomaly Detection Techniques and Applications
- Advanced Vision and Imaging
- Advanced Image and Video Retrieval Techniques
- Advanced Neural Network Applications
- Remote Sensing and Land Use
- Image Enhancement Techniques
- Network Security and Intrusion Detection
- Generative Adversarial Networks and Image Synthesis
- Model Reduction and Neural Networks
- Face recognition and analysis
- Collaboration in agile enterprises
- Biometric Identification and Security
- Photoacoustic and Ultrasonic Imaging
- Human Pose and Action Recognition
- Digital Media Forensic Detection
- Machine Learning and Data Classification
- Computer Graphics and Visualization Techniques
- Face and Expression Recognition
- Wood and Agarwood Research
Yunnan University
2021-2024
Alibaba Group (China)
2022-2024
Alibaba Group (United States)
2024
University of Edinburgh
2023
Nankai University
2020
University of Chinese Academy of Sciences
2016-2017
Sanya University
2017
Centre for Quantum Computation and Communication Technology
2015-2016
University of Technology Sydney
2014-2016
Guangdong University Of Finances and Economics
2005-2016
This paper reviews the first challenge on single image super-resolution (restoration of rich details in an low resolution image) with focus proposed solutions and results. A new DIVerse 2K dataset (DIV2K) was employed. The had 6 competitions divided into 2 tracks 3 magnification factors each. Track 1 employed standard bicubic downscaling setup, while unknown operators (blur kernel decimation) but learnable through high res train images. Each competition ∽100 registered participants 20 teams...
Sparse coding has been widely applied to learning-based single image super-resolution (SR) and obtained promising performance by jointly learning effective representations for low-resolution (LR) high-resolution (HR) patch pairs. However, the resulting HR images often suffer from ringing, jaggy, blurring artifacts due strong yet ad hoc assumptions that LR representation is equal to, linear with, lies on a manifold similar or same support set as corresponding representation. Motivated success...
Single-image super-resolution restores the lost structures and textures from low-resolved images, which has achieved extensive attention research community. The top performers in this field include deep or wide convolutional neural networks, recurrent networks. However, methods enforce a single model to process all kinds of structures. A typical operation is that certain layer based on ones recovered by preceding layers, ignoring characteristics image textures. In paper, we believe...
The increasing number of 3D objects in various applications has increased the requirement for effective and efficient object retrieval methods, which attracted extensive research efforts recent years. Existing works mainly focus on how to extract features conduct matching. With applications, come from different areas. In such circumstances, becomes more important. To address this issue, we propose a multi-view method using multi-scale topic models paper. our method, multiple views are first...
Traditional classification systems rely heavily on sufficient training data with accurate labels. However, the quality of collected depends labelers, among which inexperienced labelers may exist and produce unexpected labels that degrade performance a learning system. In this paper, we investigate multiclass problem where certain amount examples are randomly labeled. Specifically, show issue can be formulated as label noise problem. To perform classification, employ widely used importance...
How to effectively preserve the fine-scale details of image when noises are suppressed is one great challenges faced by scholars in field noisy fusion. The traditional fusion method tends smooth structures excessively. To overcome oversmoothing issue, we develop a novel that can perform fusion, denoising, and preservation fine simultaneously. In this method, modeled as superposition coarse details. At same time, brand new strategy developed decompose input into components for further...
This paper comprehensively reviews the recent development of image deblurring, including non-blind/blind, spatially invariant/variant deblurring techniques. Indeed, these techniques share same objective inferring a latent sharp from one or several corresponding blurry images, while blind are also required to derive an accurate blur kernel. Considering critical role restoration in modern imaging systems provide high-quality images under complex environments such as motion, undesirable...
Deep neural networks have been applied to image restoration achieve the top-level performance. From a neuroscience perspective, layerwise abstraction of knowledge in deep network can, some extent, reveal mechanisms how visual cues are processed human brain. A pivotal property brain is that similar can stimulate same neuron induce neurological signals. However, conventional do not consider this property, and resulting models are, as result, unstable regarding their internal propagation. In...
Recognizing the identity of a sketched face from photograph dataset is critical yet challenging task in many applications, not least law enforcement and criminal investigations. An intelligent identification system would rely on automatic sketch synthesis photographs, thereby avoiding cost artists manually drawing sketches. However, conventional sketch-photo methods tend to generate sketches that are consistent with artists'drawing styles. Identity-specific information often overlooked,...
The introduction of depth sensors such as Microsoft Kinect have driven research in human action recognition. Human skeletal data collected from convey a significant amount information for While there has been considerable progress recognition, most existing skeleton-based approaches neglect the fact that not all body parts move during many actions, and they fail to consider ordinal positions joints. Here, motivated by an action's category is determined local joint movements, we propose...
Single image superresolution (SR) aims to construct a high-resolution version from single low-resolution (LR) image. The SR reconstruction is challenging because of the missing details in given LR Thus, it critical explore and exploit effective prior knowledge for boosting performance. In this paper, we propose novel method by exploiting both directional group sparsity gradients features similarity weight estimation. proposed approach based on two observations: 1) most sharp edges are...
Remote sensing image scene classification is an important task of remote interpretation, which has recently been well addressed by the convolutional neural network owing to its powerful learning ability. However, due multiple types geographical information and redundant background images, most CNN-based methods, especially those based on a single CNN model ignoring combination global local features, exhibit limited performance accurate classification. To compensate for such insufficiency, we...
Non-blind image deconvolution is an ill-posed problem. The presence of noise and band-limited blur kernels makes the solution this problem non-unique. Existing techniques produce a residual between sharp estimation that highly correlated with image, kernel, noise. In most cases, different restoration models must be constructed for levels noise, resulting in low computational efficiency or redundant model parameters. Here we aim to develop single handles types noise: general non-blind...
Remote sensing image scene classification plays a significant role in remote analysis. Aiming at the problems of large transformation and scale variation background key objects images, we propose neural architecture search (NAS) method based on attention space. The network adaptively searches convolution, pooling, operations appropriate layers. To ensure stability searching process, multistage progressive fusion is proposed, which discards useless stages, reduces burden algorithm, improves...
The performance of single image super-resolution (SISR) has been largely improved by innovative designs deep architectures. An important claim raised these is that the models have large receptive field size and strong nonlinearity. However, we are concerned about question which factor, or model depth, more critical for SISR. Towards revealing answers, in this paper, propose a strategy based on dilated convolution to investigate how two factors affect Our findings from exhaustive...
With rapid developments in cloud computing, artificial intelligence, and robotic systems, ever more complex tasks, such as space ocean exploration, are being implemented by intelligent robots. Here, we propose an underwater image enhancement scheme for visual systems. The proposed algorithm its implementation enhances outputs captured robot real time. In this scheme, pulse-coupled neural network (PCNN)-based color transfer algorithms combined to enhance the image. To avoid imbalance details...
1. ROTHENBUCHER D., LI J., SIRKIN MOK B., JU W., 2016, Ghost driver: A field study investigating the interaction between humans and driverless vehicles, 25th IEEE International Symposium on Robot Human Interactive Communication (RO-MAN), 795–802. Google Scholar
Remote sensing images contain complex backgrounds and multi-scale objects, which pose a challenging task for scene classification. The performance is highly dependent on the capacity of representation as well discriminability classifier. Although multiple models possess better properties than single model these aspects, fusion strategy key component to maximize final accuracy. In this paper, we construct novel dual-model architecture with grouping-attention-fusion improve Specifically,...
Deep neural networks are vulnerable to adversarial attacks either by examples with indistinguishable perturbations which produce incorrect predictions, or noticeable transformations that still predicted as the original label. The latter case is known Type I attack which, however, has achieved limited attention in literature. We advocate vulnerability comes from ambiguous distributions among different classes resultant feature space of model, saying appearances may present similar features....
Despite achieving exceptional performance, deep neural networks (DNNs) suffer from the harassment caused by adversarial examples, which are produced corrupting clean examples with tiny perturbations. Many powerful defense methods have been presented such as training data augmentation and input reconstruction which, however, usually rely on prior knowledge of targeted models or attacks. A example its version very similar but different high-level representations in a victim model. If we can...