Article
Mathematics, Applied
Junjun Pan, Michael K. Ng, Ye Liu, Xiongjun Zhang, Hong Yan
Summary: This paper introduces an orthogonal nonnegative Tucker decomposition (ONTD) for nonnegative tensor data and develops a convex relaxation algorithm for solving the optimization problem of ONTD. The convergence of the algorithm is proved. ONTD is applied to image data sets from real-world applications, showing the effectiveness of the proposed algorithm through numerical results.
SIAM JOURNAL ON SCIENTIFIC COMPUTING
(2021)
Article
Computer Science, Artificial Intelligence
Mateusz Gabor, Rafal Zdunek
Summary: This study proposes a novel CNN compression technique based on the hierarchical Tucker-2 tensor decomposition, which achieves a significant reduction in parameters and FLOPS with a minor drop in classification accuracy. Compared to other compression methods, the HT-2 outperforms most of them.
APPLIED SOFT COMPUTING
(2023)
Article
Computer Science, Hardware & Architecture
Jun-Gi Jang, Moonjeong Park, Jongwuk Lee, Lee Sael
Summary: This article introduces a large-scale Tucker factorization method called the Very Sparse Tucker factorization (VeST) method for sparse and accurate tensor decomposition. The proposed VeST method outputs highly sparse decomposition results by iteratively determining unimportant elements and updating the remaining elements. Experiments demonstrate that VeST outperforms competitors in terms of accuracy and scalability.
JOURNAL OF SUPERCOMPUTING
(2022)
Article
Geochemistry & Geophysics
Le Sun, Huxiang Guo
Summary: A novel blind unmixing method for hyperspectral images based on tucker tensor decomposition and L-1 regularization term was proposed, outperforming the latest method ULTRA-V in comparative experiments on two simulation datasets.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2022)
Article
Computer Science, Artificial Intelligence
Krzysztof Fonal, Rafal Zdunek
Summary: Tensor decomposition is a valuable method for multilinear feature extraction and dimensionality reduction of multiway data. In this study, a hierarchical Tucker decomposition model with single-mode preservation is proposed, along with various tensor augmentation strategies for image classification using multimodal tensor subspace analysis. Experiment results show that the proposed method outperforms well-known tensor decomposition algorithms.
Article
Computer Science, Information Systems
Yichun Qiu, Guoxu Zhou, Yu Zhang, Andrzej Cichocki
Summary: This study aims to enhance the scalability of tensor decompositions for big data analysis by proposing two algorithms that can handle huge dense tensors, with empirical evidence from extensive simulations demonstrating the validity and efficiency of the proposed algorithms.
MULTIMEDIA TOOLS AND APPLICATIONS
(2021)
Article
Engineering, Electrical & Electronic
Maxence Giraud, Vincent Itier, Remy Boyer, Yassine Zniyed, Andre L. F. de Almeida
Summary: Many signal-based applications utilize the Tucker decomposition of high-dimensional/order tensors. However, the curse of dimensionality, resulting in exponentially increasing entries, poses a challenge to the Tucker model. The Higher-Order Orthogonal Iteration (HOOI) and Higher-Order Singular Value Decomposition (HOSVD) are widely used but suffer from the same curse. In this letter, a new methodology called TRIDENT is proposed, which offers similar estimation accuracy as HOSVD but with significantly reduced computational and storage costs.
IEEE SIGNAL PROCESSING LETTERS
(2023)
Article
Computer Science, Information Systems
Xiaotong Yu, Ziyan Luo
Summary: A new method based on Tucker decomposition for sparse tensor optimization is proposed for improving background subtraction in video surveillance. By constraining the sparsity of video foreground and optimizing the low-rank characteristics of video background, the method enhances foreground detection accuracy and background estimation effectiveness.
MULTIMEDIA TOOLS AND APPLICATIONS
(2021)
Article
Computer Science, Artificial Intelligence
Zhihao Xu, Benjia Chu, Jianbo Li, Zhiqiang Lv
Summary: This study proposes a Fast Autoregressive Tensor Decomposition (FATD) algorithm for online real-time traffic flow prediction. The algorithm models and predicts historical traffic flow using Tucker decomposition and Tensor Seasonal Autoregressive Integrated Moving Average (Tensor SARIMA), and recovers the predicted traffic flow data using Inverse Tucker Decomposition, achieving reduced computational costs while maintaining high prediction accuracy.
KNOWLEDGE-BASED SYSTEMS
(2023)
Article
Engineering, Electrical & Electronic
Chengfei Shi, Zhengdong Huang, Li Wan, Tifan Xiong
Summary: This paper proposes a mixture model for tensor completion by combining logDet function with Tucker decomposition, aiming to improve the accuracy and efficiency of the algorithm.
SIGNAL IMAGE AND VIDEO PROCESSING
(2021)
Article
Mathematics, Applied
Linjian Ma, Edgar Solomonik
Summary: This paper introduces a novel family of algorithms that use perturbative corrections to optimize the quadratic optimization subproblems in CP and Tucker decomposition. The proposed pairwise perturbation algorithms are easy to control and achieve convergenceto minima that are as good as ALS.
NUMERICAL LINEAR ALGEBRA WITH APPLICATIONS
(2022)
Article
Computer Science, Artificial Intelligence
Tianheng Zhang, Jianli Zhao, Qiuxia Sun, Bin Zhang, Jianjian Chen, Maoguo Gong
Summary: Low-rank tensor completion, particularly using Tensor Train (TT) method, has shown promising results in color image recovery by capturing hidden information effectively. However, TT may destroy the original tensor structure, leading to inadequate access to global information, especially when dealing with images with significant missing data. To address this issue, a new tensor completion model that combines Tucker rank and Tensor Train rank has been proposed, showing effectiveness in numerical experiments using various tensor data.
APPLIED INTELLIGENCE
(2022)
Article
Environmental Sciences
Marzieh Zare, Mohammad Sadegh Helfroush, Kamran Kazemi, Paul Scheunders
Summary: A novel approach based on coupled non-negative tensor decomposition is proposed in this paper, which preserves the spatio-spectral structure of hyperspectral images. By directly imposing NTD on the coupled tensors of HSI and MSI, the method can represent the intrinsic spatio-spectral structure without loss and exploit spatial and spectral information interdependently.
Article
Mathematics, Applied
Liqun Qi, Yannan Chen, Mayank Bakshi, Xinzhen Zhang
Summary: A new tensor decomposition method called triple decomposition is introduced in this paper, which decomposes a third order tensor into three factor tensors with low dimensions. Experimental results show that third order tensor data from practical applications have low triple ranks.
SIAM JOURNAL ON MATRIX ANALYSIS AND APPLICATIONS
(2021)
Article
Computer Science, Software Engineering
Zisen Fang, Fumin Qi, Yichuan Dong, Yong Zhang, Shengzhong Feng
Summary: This article focuses on parallel Tucker decomposition of dense tensors on distributed-memory systems. The proposed method utilizes hierarchical SVD to accelerate the SVD step and includes a data distribution strategy. It is found that the proposed method has lower communication cost compared to state-of-the-art methods in large-scale parallel cases.
CONCURRENCY AND COMPUTATION-PRACTICE & EXPERIENCE
(2023)
Article
Engineering, Electrical & Electronic
Chunqiang Yu, Xianquan Zhang, Xinpeng Zhang, Guoxiang Li, Zhenjun Tang
Summary: This study proposes a new reversible data hiding method with hierarchical embedding, which achieves a high embedding payload through hierarchical label map generation and embedding technique. Experimental results show that the proposed method outperforms other methods in terms of payload.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2022)
Article
Computer Science, Information Systems
Binfei Chu, Yiting Lin, Bineng Zhong, Zhenjun Tang, Xianxian Li, Jing Wang
Summary: This article proposes a novel occluder-aware representation learning framework to improve tracking performance by localizing occluders and guiding representation learning. Experimental results show that this method achieves state-of-the-art performance in multiple benchmark tests.
ACM TRANSACTIONS ON MULTIMEDIA COMPUTING COMMUNICATIONS AND APPLICATIONS
(2023)
Article
Engineering, Electrical & Electronic
Mengzhu Yu, Zhenjun Tang, Xianquan Zhang, Bineng Zhong, Xinpeng Zhang
Summary: This paper proposes a perceptual hashing algorithm for reduced-reference IQ assessment using complementary color wavelet transform and compressed sensing. Experimental results demonstrate that the proposed algorithm outperforms some state-of-the-art algorithms in terms of image classification and RR IQ assessment.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2022)
Article
Computer Science, Hardware & Architecture
Ziqing Huang, Zhenjun Tang, Xianquan Zhang, Linlin Ruan, Xinpeng Zhang
Summary: Perceptual image hashing is an effective and efficient method for identifying images in large-scale databases. One of its challenges is finding the right balance between robustness and discrimination. This study proposes a novel hashing method, HLPP, which utilizes Gabor filtering and LPP to improve robustness and discrimination. The results show that HLPP outperforms state-of-the-art algorithms in benchmark databases and copy detection experiments.
IEEE TRANSACTIONS ON DEPENDABLE AND SECURE COMPUTING
(2023)
Article
Computer Science, Information Systems
Zhenjun Tang, Zhiyuan Chen, Zhixin Li, Bineng Zhong, Xianquan Zhang
Summary: Image Quality Assessment (IQA) is a critical task of computer vision. In this paper, a novel method called UniDASTN is proposed for FR-IQA, which combines the Dual-Attention and Siamese Transformer Network. The proposed method effectively evaluates the distortion in distorted images through the spatial attention module and the dual-attention strategy. The experiments on standard IQA databases demonstrate that UniDASTN outperforms some state-of-the-art FR-IQA methods.
ACM TRANSACTIONS ON MULTIMEDIA COMPUTING COMMUNICATIONS AND APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Yufei Zeng, Zhixin Li, Zhenjun Tang, Zhenbin Chen, Huifang Ma
Summary: This paper proposes a method called HIS-MSA for multimodal sentiment analysis, which addresses the challenges of fully leveraging domain-specific knowledge and lacking effective integration methods by utilizing in-domain self-supervised multi-task learning and heterogeneous graph convolution. Experimental results demonstrate an average improvement of approximately 1.5 points in all metrics compared to the current state-of-the-art model.
EXPERT SYSTEMS WITH APPLICATIONS
(2023)
Article
Engineering, Electrical & Electronic
Yaozong Zheng, Bineng Zhong, Qihua Liang, Zhenjun Tang, Rongrong Ji, Xianxian Li
Summary: Despite the importance of both local and context information for robust tracking, existing methods in the field of CNN and Transformer mainly focus on one aspect. To address this issue, we propose the SiamPIN tracker, which consists of two modules: Global Aggregation Block (GAB) and Local Process Block (LPB). GAB captures the global context using a transformer-based architecture, while LPB extracts local information using a CNN model. These modules are connected consecutively to enable the interaction of global-local information flow. The proposed tracker achieves state-of-the-art performance on benchmark datasets while maintaining real-time running speed.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2023)
Article
Engineering, Electrical & Electronic
Jie Ma, Xiangyuan Lan, Bineng Zhong, Guorong Li, Zhenjun Tang, Xianxian Li, Rongrong Ji
Summary: In this paper, a robust anchor-free based tracking model with uncertainty estimation is proposed. A data-driven uncertainty estimation strategy is used to generate uncertainty-aware features. A pyramid-wise cross correlation operation is constructed to extract multi-scale semantic features for enhancing tracking robustness. Experimental results show that our tracker achieves competitive performance with a frame rate of 130 FPS.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2023)
Article
Computer Science, Artificial Intelligence
Xiaoping Liang, Zhenjun Tang, Ziqing Huang, Xianquan Zhang, Shichao Zhang
Summary: This paper proposes an efficient hashing method for image copy detection using 2D-2D PCA. By combining PCT and 2D-2D PCA, rotation-invariant low-dimensional features are extracted, and vector distances are used for robust and compact hash construction. Experimental results demonstrate that the proposed method outperforms representative hashing methods in classification and copy detection performance.
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING
(2023)
Article
Computer Science, Artificial Intelligence
Zhe Chen, Xiao-Jun Wu, Josef Kittler
Summary: This paper proposes a Fisher regularized e-dragging framework for image classification, which improves the intraclass compactness and interclass separability of relaxed labels. The Fisher criterion and e-dragging technique are integrated into a unified model, achieving superior performance compared to other classification methods.
IEEE TRANSACTIONS ON COGNITIVE AND DEVELOPMENTAL SYSTEMS
(2023)
Article
Engineering, Electrical & Electronic
Chunqiang Yu, Xianquan Zhang, Chuan Qin, Zhenjun Tang
Summary: This study proposes a new reversible data hiding method using Chinese remainder theorem-based secret sharing and hybrid coding, which achieves high embedding capacity and good security properties.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2023)
Article
Computer Science, Hardware & Architecture
Zhen Chen, Di Zhou, Enrico Zio, Tangbin Xia, Ershun Pan
Summary: Degradation modeling and prognostics are important for system health management. This study proposes a novel feature fusion-based HI construction method with deep learning and multiobjective optimization. Multiple degradation features are fused by a deep neural network (DNN) and several desired properties for prognostics are used to formulate the objective functions of DNN training. A multiobjective optimization model is generated to balance the complexity and performance of the fusion model. The proposed method is illustrated with two cases to demonstrate its effectiveness and robustness.
IEEE TRANSACTIONS ON RELIABILITY
(2023)
Article
Computer Science, Information Systems
Xiaoping Liang, Zhenjun Tang, Jingli Wu, Zhixin Li, Xinpeng Zhang
Summary: This paper proposes a robust image hashing algorithm based on Isometric Mapping and saliency map for copy detection, which outperforms state-of-the-art algorithms in classification and copy detection performance.
IEEE TRANSACTIONS ON MULTIMEDIA
(2023)
Article
Engineering, Electrical & Electronic
Yu Dong, Xianquan Zhang, Chunqiang Yu, Zhenjun Tang
Summary: Digital images can easily be corrupted during transmission. Most existing image denoising methods fail to effectively restore the secret image extracted from a corrupted stego image. To address this issue, a new method based on convex hull and elite opposition-based learning strategy is proposed. The method calculates pixel distortion values of the corrupted secret image and classifies them into trustable and untrusted pixels. Trustable pixels within a convex hull restore the untrusted pixel in the hull, and the remaining untrusted pixels are restored using an elite opposition-based learning strategy. Experimental results demonstrate that the proposed method outperforms state-of-the-art methods in terms of recovered secret image quality.
IEEE SIGNAL PROCESSING LETTERS
(2023)
Article
Computer Science, Artificial Intelligence
Zedu Chen, Bineng Zhong, Guorong Li, Shengping Zhang, Rongrong Ji, Zhenjun Tang, Xianxian Li
Summary: SiamBAN is a simple and effective tracker that predicts target boxes in a per-pixel fashion through a fully convolutional network, addressing the challenge of variation in scales or aspect ratios. It divides the tracking problem into classification and regression tasks to handle inconsistency, and achieves promising performance in benchmark tests.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2023)