Article
Engineering, Electrical & Electronic
Zhiyuan Zhang, Hongjin He
Summary: The paper proposes a cartoon-texture image decomposition model based on low-rank texture prior, which is able to perform perfectly on globally well-patterned images.
SIGNAL PROCESSING-IMAGE COMMUNICATION
(2021)
Article
Operations Research & Management Science
Laura Antonelli, Valentina De Simone, Marco Viola
Summary: In this study, a novel image segmentation model is proposed that can accurately handle images containing noise or oscillatory information such as texture, in addition to smooth images.
COMPUTATIONAL OPTIMIZATION AND APPLICATIONS
(2023)
Article
Computer Science, Information Systems
Jianlou Xu, Wanqing Shang, Yuying Guo
Summary: In this paper, a new cartoon-texture decomposition model with divergence-free vector field constraint is proposed, which improves the accuracy and completeness of the texture by considering the directional information lost in the existing partial decomposition models. The constrained optimization problem is transformed into an unconstrained problem and solved with the augmented Lagrange method and the alternating direction method. Experimental results confirm the effectiveness of the proposed model.
MULTIMEDIA TOOLS AND APPLICATIONS
(2023)
Article
Geochemistry & Geophysics
Xin Tian, Yuerong Chen, Changcai Yang, Jiayi Ma
Summary: The study proposes a variational pansharpening method by exploiting cartoon-texture similarities, allowing the fused high-spatial resolution MS image to preserve both global and local spatial details effectively. By leveraging the similarities of cartoon and texture components from PAN and MS images, the proposed method outperforms state-of-the-art pansharpening methods in terms of both visual effect and objective metrics, as demonstrated through extensive experiments on satellite data sets and a simulated vegetation coverage change experiment.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2022)
Article
Engineering, Electrical & Electronic
Jianlou Xu, Wanqing Shang, Yan Hao
Summary: This paper introduces a method for decomposing an image into structural and oscillatory components. By placing the structural component in the bounded variation space and the oscillatory texture in the Sobolev space, with the residual part modeled by the H-1 norm, the new model effectively decomposes the image while preserving some edges and contours.
SIGNAL IMAGE AND VIDEO PROCESSING
(2022)
Article
Mathematics, Applied
Jie Lin, Ting-Zhu Huang, Xi-Le Zhao, Tian-Hui Ma, Tai-Xiang Jiang, Yu-Bang Zheng
Summary: In this paper, a non-convex low-rank tensor approximation (NonLRTA) model is proposed for mixed noise removal in remote sensing hyperspectral images. An efficient augmented Lagrange multiplier (ALM) algorithm is developed to solve the proposed model. Experiments validate the superiority of the proposed method compared to state-of-the-art matrix-based and tensor-based methods.
APPLIED MATHEMATICS AND COMPUTATION
(2021)
Article
Mathematics, Applied
Riya Ruhela, Bhupendra Gupta, Subir Singh Lamba
Summary: Decomposition of an image into cartoon and texture components is commonly used in image processing. However, obtaining the texture component is challenging due to its varying nature. This paper introduces a new non-convex surrogate of the rank that assigns different weights to each singular value, providing an efficient way to extract both cartoon and texture components. The proposed model performs well on image restoration problems and works effectively on globally patterned and natural images.
COMPUTERS & MATHEMATICS WITH APPLICATIONS
(2022)
Article
Mathematics, Applied
You-Wei Wen, Mingchao Zhao, Michael Ng
Summary: The Meyer model has been extended to color images in opponent color space, introducing L-1 and L-infinity norms for vector-valued vectors, and redefining TV and G-norms accordingly. Dual formulations are used to handle the non-differentiability of norms, and a first-order primal-dual algorithm is applied to compute the saddle point of the minimax problem for decomposition of color images. The proposed model shows promising performance in numerical results.
APPLIED MATHEMATICS AND COMPUTATION
(2022)
Article
Engineering, Electrical & Electronic
Baoshun Shi, Chunzi Zhu, Lingyan Li, Huagui Huang
Summary: This paper investigates the task of recovering normal-exposure images from low-light images and proposes a cartoon-texture guided network called CatNet. By utilizing a cartoon-guided normalizing flow and an elaborated frequency domain attention mechanism, CatNet is able to enhance images while preserving more details and richer colors.
DIGITAL SIGNAL PROCESSING
(2024)
Article
Materials Science, Textiles
Runhu Zhu, Binjie Xin, Na Deng, Mingzhu Fan
Summary: This article proposes a new fabric defect detection method based on the latest cartoon texture image decomposition model, visual saliency algorithm, and mathematical morphology. Experiments conducted using a self-made dataset and comparison with other common fabric defect detection methods show that this method has high detection accuracy and efficiency, and outperforms other methods in subjective vision and objective evaluation.
TEXTILE RESEARCH JOURNAL
(2023)
Article
Biology
Mohammad Jalali, Hamid Behnam, Maryam Shojaeifard
Summary: In this work, a method is proposed to reduce speckle artifacts in cardiac ultrasound imaging by decomposing the images into cartoon and texture components using convolutional sparse coding. This approach enhances image quality and segmentation accuracy, with numerical results showing significant improvement in segmentation metrics such as Hausdorff distance and Dice similarity coefficient.
COMPUTERS IN BIOLOGY AND MEDICINE
(2021)
Article
Computer Science, Information Systems
Xiaofang Li, Weiwei Wang, Xiangchu Feng, Tingting Qi
Summary: Cartoon-Texture decomposition is a fundamental task with wide applications in image processing and computer vision. Existing models introduce correlation terms to improve the separation of cartoon and texture, but they often overlook local geometric information, resulting in insufficient decorrelation. In this work, we propose a patch-wise cosine similarity for decorrelating cartoon and texture, which takes into account the local geometric information and achieves better separation. By combining this decorrelation term with regularities for cartoon and texture, we present a new and improved Cartoon-Texture decomposition model. Experimental results show that our model outperforms existing methods, especially in preserving cartoon edges.
JOURNAL OF VISUAL COMMUNICATION AND IMAGE REPRESENTATION
(2023)
Article
Computer Science, Information Systems
Xin Sheng, Jing Yuan, Wenbing Tao, Bo Tao, Liman Liu
Summary: The proposed mesh-based continuous max-flow approach improves visual quality and computational efficiency in large-scale texture mapping for 3D scene reconstruction. It utilizes MRF and Potts models to mathematically formulate view selection problems and effectively solve challenging combinatorial optimization problems. The approach defines criteria for evaluating texture quality using visual effects and partitions large 3D triangular meshes to reduce memory consumption.
INFORMATION SCIENCES
(2021)
Article
Computer Science, Artificial Intelligence
Le Sun, Chengxun He, Yuhui Zheng, Zebin Wu, Byeungwoo Jeon
Summary: In this paper, we proposed a unified subspace low-rank learning method called STCR, which can effectively exploit the low-rank nature of hyperspectral images in different domains for various low-level vision tasks.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2023)
Article
Computer Science, Information Systems
Wanqing Shang, Jianlou Xu, Yuying Guo
Summary: This paper proposes a curvature-guided model with divergence-free constraints for image decomposition. By introducing a curvature term and edge indicator function to balance edge features and smoothness, the model preserves edges and protects textures. Experiment results demonstrate the effectiveness of the proposed model.
Proceedings Paper
Acoustics
Kazuki Naganuma, Shunsuke Ono
Summary: We propose a diagonal preconditioning method for automatically selecting the step sizes of a primal-dual splitting method. This method resolves the limitations of the conventional method and establishes a practical preconditioning method. The proposed preconditioners eliminate the need for the matrix representations of linear operators and allow for computable proximity operators.
2022 30TH EUROPEAN SIGNAL PROCESSING CONFERENCE (EUSIPCO 2022)
(2022)
Article
Geochemistry & Geophysics
Saori Takeyama, Shunsuke Ono
Summary: This article proposes a new method for estimating high spatial resolution hyperspectral images based on convex optimization. The method can simultaneously estimate a high spatial resolution image and a noiseless guide image, and effectively utilize prior knowledge and spatial detail information to improve the estimation accuracy.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2022)
Article
Geochemistry & Geophysics
Shingo Takemoto, Kazuki Naganuma, Shunsuke Ono
Summary: The spatio-spectral total variation (SSTV) model is widely used for regularization of hyperspectral images (HSIs), but it struggles with removing noise while preserving complex spatial structures. To address this issue, we propose a new regularization method called graph-SSTV (GSSTV), which explicitly reflects the spatial structure of the HSI and incorporates a weighted spatial difference operator. We formulate the noise removal problem as a convex optimization problem involving GSSTV and develop an efficient algorithm to solve it.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2022)
Article
Engineering, Electrical & Electronic
Ruiki Kobayashi, Genki Fujii, Yuta Yoshida, Takeru Ota, Fumiaki Nin, Hiroshi Hibino, Samuel Choi, Shunsuke Ono, Shogo Muramatsu
Summary: A novel restoration model is proposed for optical coherence tomography (OCT) data, which tackles the challenge of weak reflected light and identifies tomographic structures using image processing and an algorithm derived from a primal-dual splitting framework. The significance of this method is verified through simulations and experiments, highlighting its importance in restoring OCT data.
IEEE TRANSACTIONS ON COMPUTATIONAL IMAGING
(2022)
Article
Geochemistry & Geophysics
Kazuki Naganuma, Shunsuke Ono
Summary: This article proposes a general destriping framework using a newly introduced stripe noise characterization named flatness constraint (FC) to handle various image regularizations. The framework formulates the destriping problem as a nonsmooth convex optimization problem involving a general form of image regularization and the FC, effectively removing stripe noise from images by mathematically modeling the constraint.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2022)
Proceedings Paper
Computer Science, Information Systems
Eisuke Yamagata, Shunsuke Ono
Summary: This paper proposes a novel recovering framework for dynamic graph signal models that leverage both temporal and vertex-domain priors, by introducing regularization terms in a convex optimization problem to capture behaviors of graph signals in the two domains and integrate the dynamics of the dynamic graph topology. Experimental comparisons with conventional frameworks on synthetic datasets demonstrate the advantageous results of the proposed method in numerous settings.
2021 ASIA-PACIFIC SIGNAL AND INFORMATION PROCESSING ASSOCIATION ANNUAL SUMMIT AND CONFERENCE (APSIPA ASC)
(2021)
Proceedings Paper
Acoustics
Kazuki Naganuma, Saori Takeyama, Shunsuke Ono
Summary: This paper introduces an effective destriping method for remote-sensing data by formulating the problem as a convex optimization problem with zero-gradient constraints. The method fully captures the nature of stripe noise and utilizes a simple operation to develop an efficient algorithm for solving the problem. The advantages of the method are demonstrated through destriping experiments using remote-sensing data.
2021 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING (ICASSP 2021)
(2021)
Proceedings Paper
Acoustics
Junya Hara, Koki Yamada, Shunsuke Ono, Yuichi Tanaka
Summary: This paper proposes a design method for sampling matrices of graph signals that ensures perfect recovery for arbitrary graph signal subspaces. Experimental results show that the proposed sampling matrix provides better reconstruction accuracy for various signal models.
2021 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING (ICASSP 2021)
(2021)
Article
Engineering, Electrical & Electronic
Takayuki Nagata, Taku Nonomura, Kumi Nakai, Keigo Yamada, Yuji Saito, Shunsuke Ono
Summary: This study proposes a sensor selection method based on the proximal splitting algorithm and the ADMM algorithm, showing better performance than existing greedy and convex relaxation methods in terms of the A-optimality criterion. The proposed method requires longer computational time than the greedy method but is shorter than the convex relaxation method in large-scale problems.
IEEE SENSORS JOURNAL
(2021)
Article
Engineering, Electrical & Electronic
Seisuke Kyochi, Shunsuke Ono, Ivan Selesnick
Summary: This paper introduces an epigraphical relaxation (ERx) technique for non-proximable mixed norm minimization, providing a method to handle a wide range of non-proximable mixed norms in optimization without changing the minimizer of the original problem. Additionally, novel regularizers based on ERx are developed and shown to be effective through experiments.
IEEE TRANSACTIONS ON SIGNAL PROCESSING
(2021)
Proceedings Paper
Acoustics
Saori Takeyama, Tatsuya Komatsu, Koichi Miyazaki, Masahito Togami, Shunsuke Ono
Summary: This paper proposes a robust acoustic scene classification (ASC) method for multiple devices using maximum classifier discrepancy (MCD) and knowledge distillation (KD), which employs domain adaptation to train device-specific ASC models and combines them into a multi-domain ASC model. The proposed method aligns class distributions using MCD for domain adaptation and improves ASC accuracy for both target and non-target devices.
28TH EUROPEAN SIGNAL PROCESSING CONFERENCE (EUSIPCO 2020)
(2021)
Article
Engineering, Electrical & Electronic
Taku Nonomura, Shunsuke Ono, Kumi Nakai, Yuji Saito
Summary: This study proposes randomized subspace Newton convex methods for sensor selection problem, where a customized approach is used to select update variables for improved performance. The results show that the randomized subspace Newton methods yield almost identical results in the converged solution compared to the original method. Furthermore, while requiring more computational steps, the randomized subspace Newton methods reduce total computational time significantly.
IEEE SIGNAL PROCESSING LETTERS
(2021)
Proceedings Paper
Computer Science, Artificial Intelligence
Saori Takeyama, Shunsuke Ono
IGARSS 2020 - 2020 IEEE INTERNATIONAL GEOSCIENCE AND REMOTE SENSING SYMPOSIUM
(2020)
Proceedings Paper
Imaging Science & Photographic Technology
Saori Takeyama, Shunsuke Ono
2020 IEEE INTERNATIONAL CONFERENCE ON IMAGE PROCESSING (ICIP)
(2020)
Proceedings Paper
Acoustics
Seisuke Kyochi, Shunsuke Ono, Ivan Selesnick
2020 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH, AND SIGNAL PROCESSING
(2020)