Article
Chemistry, Analytical
Ling Zhang, Xuefei Yang, Zhenlong Wan, Dingxin Cao, Yingcheng Lin
Summary: This paper proposes a multi-scale FPGA-based image fusion technology that significantly enhances visual enhancement capability and fusion speed. The method decomposes the source images into different layers, constructs fusion weight maps using attention mechanism, and uses fusion strategies for different layers. Image enhancement technology is also incorporated to improve contrast. The method is tested on a specific board and achieves real-time fusion with high frame rates.
Article
Optics
Yingcheng Lin, Dingxin Cao, Xichuan Zhou
Summary: This paper proposes an adaptive image fusion method, which decomposes and fuses images using rolling guidance filter and saliency detection. It assigns weights to the fused image based on the perception of significant information by the human visual system. The proposed method enhances the quality of image fusion by improving contrast and retaining details.
Article
Instruments & Instrumentation
Jianming Zhang, Wenxin Lei, Shuyang Li, Zongping Li, Xudong Li
Summary: This paper proposes a novel algorithm for infrared and visible image fusion. The algorithm decomposes the input image into different layers using a guided filter, and then fuses them using entropy-based fusion module, maximum absolute value rule, and a mask-guided deep convolutional neural network. Experimental results demonstrate that the algorithm achieves good performance in both subjective and objective evaluation.
INFRARED PHYSICS & TECHNOLOGY
(2023)
Article
Computer Science, Artificial Intelligence
Wenxia Yin, Kangjian He, Dan Xu, Yueying Luo, Jian Gong
Summary: The proposed adaptive enhanced infrared and visible image fusion algorithm can better preserve valuable texture information and prominent infrared targets, which is beneficial for target detection and tracking tasks.
NEURAL COMPUTING & APPLICATIONS
(2022)
Article
Computer Science, Interdisciplinary Applications
Zhemin Zhuang, Zengbiao Yang, Alex Noel Joseph Raj, Chuliang Wei, Pengcheng Jin, Shuxin Zhuang
Summary: This study proposed a reliable classification method for breast ultrasound images of tumors by combining image decomposition and fusion techniques with adaptive spatial feature fusion technology. The experimental results showed excellent performance in various metrics. The research has the potential to automate breast cancer detection and has strong clinical significance.
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
(2021)
Article
Engineering, Electrical & Electronic
Ming Lu, Min Jiang, Jun Kong, Xuefeng Tao
Summary: This paper proposes a real-time end-to-end visible-infrared image fusion model based on layer decomposition and re-parameterization. The model decomposes the infrared image into structural and fuzzy layers and fuses them using a re-parameterization fusion network. Experimental results demonstrate that the proposed method achieves comparable performance to the state of the art in terms of visual effect and quantitative metrics.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2023)
Article
Computer Science, Interdisciplinary Applications
Ali Ozbek, Xose Luis Dean-Ben, Daniel Razansky
Summary: This paper presents a fast adaptive OAT data compression framework operating on fully sampled tomographic data. The framework is based on a wavelet packet transform that maximizes the data compression ratio according to the desired signal energy loss. A dedicated reconstruction method efficiently renders images directly from the compressed data.
IEEE TRANSACTIONS ON MEDICAL IMAGING
(2022)
Article
Computer Science, Artificial Intelligence
Wang Cai, Ping Jiang, LeShi Shu, ShaoNing Geng, Qi Zhou
Summary: This paper introduces an innovative deep learning-based monitoring system for diagnosing the penetration state in real time during laser welding. The system captures interaction zone images using a high-speed camera and employs an adaptive fusion method to eliminate interference and highlight keyhole and molten pool. A deep learning model based on convolutional neural network is established to model the relationship between fusion images and penetration states. Validation results show high accuracy and short latency, indicating the system's suitability for real-time monitoring.
JOURNAL OF INTELLIGENT MANUFACTURING
(2023)
Article
Engineering, Electrical & Electronic
Xudong Sun, Yuan Zhu, Xianping Fu
Summary: This article proposes a novel underwater imaging method based on RGB and optimal band image fusion. It selects the most appropriate wavelength for underwater observation using a band selection algorithm based on multicriteria decision-making. The underwater stereo-imaging system with color and optimal waveband is used to collect images in different modalities, and an iteration-free fusion model based on multiscale decomposition is utilized to generate enhanced underwater images. Experimental results show that the images acquired by this system have richer details and more prominent objects compared to those captured by an RGB camera.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2023)
Article
Computer Science, Software Engineering
D. Lin, L. Seiler, C. Yuksel
Summary: The paper presents a unified refactoring of quadratic and cubic interpolations as standard linear interpolation plus linear interpolations of higher-order terms, showing their effectiveness in various applications. The formulations provided can significantly reduce computation cost compared to typical higher-order interpolations and allow for adaptive higher-order interpolation as needed. Additionally, the authors describe how minor modifications to existing GPU hardware could provide hardware support for quadratic and cubic interpolations.
COMPUTER GRAPHICS FORUM
(2021)
Article
Computer Science, Information Systems
Yanyan Liu, Changcheng Pan, Minglin Bie, Jin Li
Summary: This study proposes an efficient real-time target tracking method based on low-dimensional adaptive feature fusion. It improves tracking accuracy by adaptively fusing color and HOG features, reduces overfitting using dimension reduction, and ensures tracking accuracy with the average correlation energy estimation method. Experimental results confirm its superiority over other algorithms, achieving the highest tracking accuracy and success tracking rate on the OTB100 dataset. The proposed method also demonstrates real-time target tracking capabilities with 50+fps.
JOURNAL OF VISUAL COMMUNICATION AND IMAGE REPRESENTATION
(2022)
Article
Engineering, Electrical & Electronic
Ronghao Pei, Weiwei Fu, Kang Yao, Tianli Zheng, Shangshang Ding, Hetong Zhang, Yang Zhang
Summary: This paper proposes a new method for multi-focus microscopic image fusion, which utilizes the improved SegNet network and a parallel fusion strategy to quickly generate full-focus images.
IEEE PHOTONICS JOURNAL
(2021)
Article
Biology
Zhenzhong Liu, Laiwang Zheng, Lin Gu, Shubin Yang, Zichen Zhong, Guobin Zhang
Summary: In robot-assisted surgery, precise surgical instrument segmentation technology plays a crucial role in facilitating efficient and safe surgical operations. This article introduces an effective surgical instrument segmentation network called InstrumentNet, which utilizes YOLOv7 as the object detection framework to achieve real-time detection. Experimental results demonstrate that the proposed model achieves excellent segmentation performance on surgical instruments compared to other advanced models, highlighting its universality and superiority.
COMPUTERS IN BIOLOGY AND MEDICINE
(2023)
Article
Computer Science, Artificial Intelligence
Hegui Zhu, Kai Wang, Ziwei Zhang, Yuelin Liu, Wuming Jiang
Summary: This article introduces a new method for improving the quality of low-light images and obtaining clear images. By decomposing the image and using adaptive information fusion, it successfully restores brightness, removes noise, and achieves good performance in experiments.
NEURAL COMPUTING & APPLICATIONS
(2022)
Article
Engineering, Biomedical
Yang Zheng, Xiaogang Hu
Summary: An adaptive real-time decomposition approach has been developed for prolonged muscle activation. It increases the identifiable motor unit (MU) number and improves decomposition accuracy by periodically optimizing and updating the separation matrix. This approach allows for longitudinal evaluation of MU firing and recruitment properties and enhances neural decoding performance.
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING
(2022)
Article
Automation & Control Systems
Jiayi Ma, Kaining Zhang, Junjun Jiang
Summary: A novel appearance-based loop closure detection system is proposed in this work, which selects candidate frames using Super-features and ASMK, and verifies loops using LPM-GC algorithm. Experimental results demonstrate that the proposed method achieves good performance in loop closure detection task.
IEEE-CAA JOURNAL OF AUTOMATICA SINICA
(2023)
Article
Automation & Control Systems
Xin Tian, Wei Zhang, Dian Yu, Jiayi Ma
Summary: This letter proposes a new hyperspectral fusion paradigm that simultaneously fuses hyperspectral, multispectral, and panchromatic images. It introduces a novel sparse tensor prior using patch-based sparse tensor dictionary learning to better describe the inherent structures of high-resolution hyperspectral images. The effectiveness of the proposed method is validated through numerous experiments in terms of visual quality and quantitative analysis.
IEEE-CAA JOURNAL OF AUTOMATICA SINICA
(2023)
Article
Computer Science, Artificial Intelligence
Ziwei Shi, Guobao Xiao, Linxin Zheng, Jiayi Ma, Riqing Chen
Summary: In this paper, a novel Joint Representation Attention Network (JRA-Net) is proposed to establish reliable correspondences for image pairs. The attention mechanism and weight function are used to improve the reliability of correspondences and enhance the generalization ability. Empirical experiments demonstrate the effectiveness and superiority of JRA-Net.
PATTERN RECOGNITION
(2023)
Article
Computer Science, Artificial Intelligence
Yuanbin Fu, Jiayi Ma, Xiaojie Guo
Summary: This paper presents a hierarchical image organization framework based on scale-space perspective. It converts the original complex problem into a series of two-component separation sub-problems, and provides theoretical and experimental results to demonstrate its effectiveness and superiority.
COMPUTER VISION AND IMAGE UNDERSTANDING
(2023)
Article
Computer Science, Artificial Intelligence
Xingyu Jiang, Shihua Zhang, Xiao-Ping Zhang, Jiayi Ma
Summary: This paper proposes an enhanced sparse GNN method using Guided Attentional Pooling to emphasize informative cues in GNN layers. It extracts information-rich keypoints and uses them to guide attentional pooling for better information preservation. Experimental results show that our method outperforms existing techniques in tasks such as camera pose estimation, fundamental matrix estimation, and visual localization.
COMPUTER VISION AND IMAGE UNDERSTANDING
(2023)
Article
Automation & Control Systems
Xinyu Ye, Jiayi Ma
Summary: This article proposes an effective and efficient visual place recognition (VPR) approach that integrates semantic, sequential, and spatial geometric information. The focus is on candidate selection and geometric verification, rather than feature extraction. The proposed method, neighborhood manifold preserving matching (NMP), utilizes sequence partitioning and sequence-to-sequence matching to improve VPR performance. Experimental results demonstrate the superiority of the proposed VPR method and its potential for integration with other pipelines.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2023)
Article
Computer Science, Artificial Intelligence
Erting Pan, Yong Ma, Xiaoguang Mei, Fan Fan, Jiayi Ma
Summary: Hyperspectral image denoising is a challenging problem, and prior knowledge about hyperspectral noise is essential for developing an effective denoising method. Most existing methods assume equal noise intensity in all bands, which contradicts practical HSIs and leads to unsatisfactory results. To address this, we propose a novel denoising framework called (N) over cap-Net, which utilizes the intrinsic properties of real HSI noise and employs a bootstrap mechanism for better denoising performance.
PATTERN RECOGNITION
(2023)
Article
Computer Science, Artificial Intelligence
Hao Liang, Rui Liu, Zhongyuan Wang, Jiayi Ma, Xin Tian
Summary: Deep learning-based approaches have achieved significant results in Poisson denoising under low-light conditions. However, most existing methods focus on network architecture design without physical interpretability, making them unsuitable for blind denoising in real environments. To address this, the authors propose VBDNet, a variational Bayesian deep network that combines Bayesian inference and deep learning for blind Poisson denoising. VBDNet outperforms state-of-the-art methods on synthetic and natural data.
PATTERN RECOGNITION
(2023)
Article
Geochemistry & Geophysics
Chengjie Ke, Wei Zhang, Zhongyuan Wang, Jiayi Ma, Xin Tian
Summary: We propose a coarse-to-fine adaptation learning fusion network for pansharpening, which combines the advantages of UNet and Transformer architectures to explore texture information of different characteristics. The network adjusts the spatial and spectral information of the coarse-fusion image based on target-specific knowledge in an unsupervised learning manner. Experiments show that our proposed method outperforms other state-of-the-art DL methods in terms of visual quality and quantitative analysis.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Computer Science, Artificial Intelligence
Erting Pan, Yong Ma, Xiaoguang Mei, Jun Huang, Qihai Chen, Jiayi Ma
Summary: In this article, a novel approach for denoising and destriping HSI is proposed. By decomposing the task into auxiliary sub-tasks, the shortcomings of the generalized mathematical model are addressed, leading to accurate destriping and high-fidelity restoration.
PATTERN RECOGNITION
(2023)
Article
Computer Science, Artificial Intelligence
Qing Ma, Junjun Jiang, Xianming Liu, Jiayi Ma
Summary: To address the problem of hyperspectral image super-resolution, this paper proposes a novel method that uses the Transformer architecture to learn the prior of hyperspectral images (HSIs). By adding a 3D-CNN behind the Transformer layers, the method aims to capture the spatio-spectral correlation of HSIs. Experimental results demonstrate that the proposed method outperforms existing algorithms in terms of quantitative and visual quality.
INFORMATION FUSION
(2023)
Article
Geochemistry & Geophysics
Erting Pan, Yong Ma, Xiaoguang Mei, Fan Fan, Jun Huang, Jiayi Ma
Summary: This study proposes a progressive hyperspectral destriping method with an adaptive frequency focus for accurate destriping and delicate restoration. The method encodes the degraded input to the frequency domain with smaller scales and separates noise and preserves details in the high-frequency domain. The experimental results demonstrate the superiority of the proposed method over the current state-of-the-art methods.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Geochemistry & Geophysics
Meiqi Gong, Hao Zhang, Han Xu, Xin Tian, Jiayi Ma
Summary: In this article, a novel multipatch and multistage pansharpening method called PSDNet is proposed, which utilizes knowledge distillation. The method incorporates multipatch inputs and a multistage network for more accurate learning. Small patches are used in the early part to learn accurate local information, while large patches are employed later to fine-tune for overall information. The multistage network reduces the difficulty of single-step pansharpening and generates elaborate results progressively. Distillation loss is introduced to reinforce the guidance of the ground truth, leading to superior performance compared to existing state-of-the-art methods. The code for PSDNet is available at https://github.com/Meiqi-Gong/PSDNet.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Computer Science, Artificial Intelligence
Hebaixu Wang, Hao Zhang, Xin Tian, Jiayi Ma
Summary: Pansharpening is a technique that combines a high-resolution panchromatic image and a low-resolution multi-spectral image to generate a high-resolution multi-spectral image. The proposed Zero-Sharpen method, which combines deep learning and variational optimization, can be easily applied across different satellites and reduces scale variance. Extensive experiments demonstrate the superiority of this method over existing ones.
INFORMATION FUSION
(2024)
Article
Computer Science, Theory & Methods
Junjun Jiang, Chenyang Wang, Xianming Liu, Jiayi Ma
Summary: This survey systematically reviews deep learning-based face super-resolution (FSR) methods. It summarizes the problem formulation of FSR, introduces assessment metrics and loss functions. It elaborates on facial characteristics and popular datasets used in FSR, and categorizes existing methods based on the utilization of facial characteristics. For each category, it provides a general description of design principles, an overview of representative approaches, and discusses their pros and cons. The survey also evaluates the performance of state-of-the-art methods and introduces joint FSR and other tasks, as well as FSR-related applications, while envisioning future technological advancements in this field.
ACM COMPUTING SURVEYS
(2023)
