A5100433758- Advanced Image Processing Techniques
- Image and Signal Denoising Methods
- Image Processing Techniques and Applications
- Advanced Vision and Imaging
- Medical Imaging Techniques and Applications
- Generative Adversarial Networks and Image Synthesis
- Advanced Neural Network Applications
- Smart Agriculture and AI
- Advanced Radiotherapy Techniques
- Remote-Sensing Image Classification
- Remote Sensing in Agriculture
- Advanced Data Compression Techniques
- Brain Tumor Detection and Classification
- CCD and CMOS Imaging Sensors
- Sparse and Compressive Sensing Techniques
- Human Pose and Action Recognition
- Cell Image Analysis Techniques
- Image Enhancement Techniques
- Digital Holography and Microscopy
Aerospace Information Research Institute
2022-2025
Chinese Academy of Sciences
2022-2025
University of Chinese Academy of Sciences
2022-2024
Southwest Jiaotong University
2019-2023
Sichuan University
2023
West China Hospital of Sichuan University
2023
Beihang University
2023
Southeast University
2021
Tsinghua University
2020
Southern Medical University
2013-2017
Recovering sharp video sequence from a motion-blurred image is highly ill-posed due to the significant loss of motion information in blurring process. For event-based cameras, however, fast can be captured as events at high frame rate, raising new opportunities exploring effective solutions. In this paper, we start sequential formulation deblurring, then show how its optimization unfolded with novel end-toend deep architecture. The proposed architecture convolutional recurrent neural network...
This paper reviews the video colorization challenge on New Trends in Image Restoration and Enhancement (NTIRE) workshop, held conjunction with CVPR 2023. The target of this is converting grayscale videos into color better performance temporal consistency. consists two tracks. For Track 1, goal achieving best FID (Fréchet Inception Distance) while being constrained to maintain or improve over baseline method terms temporal-consistency metric. Color Distribution Consistency (CDC) index used as...
Arable land is fundamental to agricultural production and a crucial component of ecosystems. However, its complex texture distribution in remote sensing images make it susceptible interference from other cover types, such as water bodies, roads, buildings, complicating accurate identification. Building on previous research, this study proposes an efficient lightweight CNN-based network, U-MGA, address the challenges feature similarity between arable non-arable areas, insufficient...
4D-CT plays an important role in lung cancer treatment because of its capability providing a comprehensive characterization respiratory motion for high-precision radiation therapy. However, due to the inherent high-dose exposure associated with CT, dense sampling along superior-inferior direction is often not practical, thus resulting inter-slice thickness that much greater than in-plane voxel resolutions. As consequence, artifacts such as vessel discontinuity and partial volume effects are...
The convolution neural networks (CNNs) are widely used in computer vision applications nowadays. However, the trends of higher accuracy and resolution generate larger networks, indicating that computation I/O bandwidth key bottlenecks to reach performance. Xilinx's latest 7nm Versal ACAP platform with AI-Engine (AIE) cores can deliver up-to 8x silicon compute density at 50% power consumption compared traditional FPGA solutions. In this paper, we propose XVDPU: AIE-based int8-precision CNN...
Motion blur recovery is a common method in the field of remote sensing image processing that can effectively improve accuracy detection and recognition. Among existing motion methods, algorithms based on deep learning do not rely priori knowledge and, thus, have better generalizability. However, usually suffer from feature misalignment, resulting high probability missing details or errors recovered images. This paper proposes an end-to-end generative adversarial network (SDD-GAN) for...
A sparse camera array utilizes several identical cameras for multi-frame super-resolution. However, the influence of layout on super-resolution performance remains unclear. In conventional arrays, all share a similar observation model, resulting in redundancy during information collection. This study presents novel framework that actively controls to rotate certain degrees. Through detailed analysis forward imaging it is shown rotation uses rotational asymmetry property pixel layouts and...
Today, convolutional neural networks (CNNs) are widely used in computer vision applications. However, the trends of higher accuracy and resolution generate larger networks. The requirements computation or I/O key bottlenecks. In this article, we propose XVDPU: AI Engine (AIE)-based CNN accelerator on Versal chips to meet heavy requirements. To resolve IO bottleneck, adopt several techniques improve data reuse reduce An arithmetic logic unit is further proposed that can better balance...
Multi-frame super-resolution (MFSR) leverages complementary information between image sequences of the same scene to increase resolution reconstructed image. As a branch MFSR, burst aims restore details by leveraging noisy sequences. In this paper, we propose an efficient burst-enhanced network (BESR). Specifically, introduce Geformer, gate-enhanced transformer, and construct enhanced CNN-Transformer block (ECTB) combining convolutions enhance local perception. ECTB efficiently aggregates...
3-Dimensional (3D) convolutional neural networks (CNN) are widely used in the field of disparity estimation. However, 3D CNN is more computationally dense than 2D due to increase dimension. To enable practical applications autonomous driving, robotics, and other scenarios on embedded devices, we propose a unified 2D/3D accelerator (A-U3D) design. This design unifies standard / transposed convolution into convolution, respectively. Our processing unit can support same mode without additional...
Although the existing deblurring methods for defocused images are capable of approximately recovering clear images, they still exhibit certain limitations, such as ringing artifacts and remaining blur. Along these lines, in this work, a novel deep-learning-based method image defocus was proposed, which can be applied to medical traffic monitoring, other fields. The developed approach is equipped with wavelet transform, an iterative filter adaptive module, graph neural network specifically...
Lung 4D computed tomography (4D-CT), which is a time-resolved CT data acquisition, performs an important role in explicitly including respiratory motion treatment planning and delivery. However, the radiation dose usually reduced at expense of inter-slice spatial resolution to minimize radiation-related health risk. Therefore, enhancement along superior-inferior direction necessary. In this paper, super-resolution (SR) reconstruction method based on patch low-rank matrix proposed improve...
Recovering sharp video sequence from a motion-blurred image is highly ill-posed due to the significant loss of motion information in blurring process. For event-based cameras, however, fast can be captured as events at high time rate, raising new opportunities exploring effective solutions. In this paper, we start sequential formulation deblurring, then show how its optimization unfolded with novel end-to-end deep architecture. The proposed architecture convolutional recurrent neural network...
Lung four-dimensional computer tomography (4D-CT) data resolution enhancement is helpful for lung cancer accurate radiotherapy. Sparse representation based algorithm has been proposed to reconstruct the image, and achieved state-of-art performance. In this paper, on sparse algorithm, we present an adaptively patch partition approach divide slices into scaled patches. This will catch more anatomical nuances improve reconstruction. The quad tree-based employed in our method slices. Moreover, a...
Recently, convolution neural network (CNN) has been widely used in single image super-resolution (SR). However, the traditional structure problems of fewer layers and slow convergence speed. In this paper, an method based on deep residual is proposed. Through deepening structure, more receptive fields are obtained. Thus, pixel information utilized to improve reconstruction accuracy model. The feature extraction process carried out directly low resolution space, images sampled by shuffling...