A5073130314- Advanced Image Processing Techniques
- Advanced Memory and Neural Computing
- Generative Adversarial Networks and Image Synthesis
- Video Analysis and Summarization
- Advanced Vision and Imaging
- Image and Signal Denoising Methods
- Advanced Image Fusion Techniques
- Music and Audio Processing
- Advanced Battery Materials and Technologies
- Supercapacitor Materials and Fabrication
- CCD and CMOS Imaging Sensors
- Speech Recognition and Synthesis
- Remote-Sensing Image Classification
- Face recognition and analysis
- Advanced Neural Network Applications
- Image Enhancement Techniques
- Human Motion and Animation
- Image and Object Detection Techniques
- Advanced MRI Techniques and Applications
- GaN-based semiconductor devices and materials
- Robotics and Sensor-Based Localization
- Industrial Vision Systems and Defect Detection
- Human Pose and Action Recognition
- Radar Systems and Signal Processing
- Indoor and Outdoor Localization Technologies
Nanyang Technological University
2022-2025
Beijing Institute of Technology
2017-2021
Beijing Institute of Optoelectronic Technology
2017
Xiamen University
2011
Xiamen University of Technology
2011
Chung Yuan Christian University
2008
Microsoft Research Asia (China)
2004-2005
Blind deblurring has received considerable attention in recent years. However, state-of-the-art methods often fail to process saturated blurry images. The main reason is that pixels around regions are not conforming the commonly used linear blur model. Pioneer arts suggest excluding these during process, which sometimes simultaneously removes informative edges and results insufficient information for kernel estimation when large exist. To address this problem, we introduce a new model fit...
The success of existing face deblurring methods based on deep neural networks is mainly due to the large model capacity. Few algorithms have been specially designed according domain knowledge images and physical properties process. In this paper, we propose an effective algorithm convolutional (CNNs). Motivated by conventional process which usually involves motion blur estimation latent clear image restoration, proposed first estimates a CNN then restores with estimated blur. However,...
Deblurring night blurry images is difficult, because the common-used blur model based on linear convolution operation does not hold in this situation due to influence of saturated pixels. In paper, we propose a non-blind deblurring network (NBDN) restore images. To mitigate side effects brought by pixels that violate model, develop confidence estimation unit (CEU) estimate map which ensures smaller contributions these deconvolution steps are optimized conjugate gradient (CG) method....
Color has been widely used in sports video analysis. Previous techniques, however require color models from prior information or user interaction, and do not address the problem of how to automatically form a an arbitrary setting. In this paper, we propose automatic technique for extracting playing surface team uniforms, which can be higher-level processes such as tracking recognition. Unlike most previous methods, our approach is capable handling multi-colored patterns like striped uniforms...
Slow-motion replays highlight important and exciting events in sports videos. Previous works for slow-motion replay detection, however, are usually limited to only some specific production rules such as frame repetition, special video effects transitions. In this paper, we present a generic method detecting based on the difference of motions between normal shots within same shot class. Experiments different types videos have verified our approach achieve reasonable results.
Most existing deep learning-based motion segmentation methods treat as a binary problem, which is generally not the real case in dynamic scenes. In addition, object and camera are often mixed, making problem difficult. This paper proposes joint learning method fuses semantic features clues using CNNs with deformable convolution embedding module, to address multi-object problem. The module serves fusion color information. And learns distinguish objects' status inspiration from geometric...