Article
Mathematics, Applied
Tatsu Kuwatani, Hideitsu Hino, Kenji Nagata, Takahiro Kawashima, Mitsuhiro Toriumi, Masato Okada
Summary: Bayesian image processing has become increasingly important in various fields of natural sciences. It uses prior knowledge and forward models to accurately estimate the physical quantities of the target. By estimating hyperparameters, the hidden physical parameters governing the processes and structure of the target and sensing systems can be determined. This paper discusses the physical meaning and mechanism of Bayesian sensing in the spatial-inversion problem and proposes a solution.
Article
Computer Science, Information Systems
Yongwei Li, Filiberto Pla, Marten Sjostrom, Ruben Fernandez-Beltran
Summary: Recent studies have shown the potential and advantages of different light field information processes. In this paper, the interaction between color interpolation and depth estimation in light field is addressed, proposing a probabilistic approach to handle these two processing steps jointly. Experimental results demonstrate that both image interpolation quality and depth estimation can benefit from their interaction.
Article
Mathematics, Applied
Juan Baz, Pedro Alonso, Juan Manuel Pena, Raul Perez-Fernandez
Summary: This paper focuses on the conditions for the covariance matrix of a multivariate Gaussian distribution to be totally positive, with a particular emphasis on Gaussian Markov Random Fields. It is proven that if the graph representing the Gaussian Markov Random Field consists of path graphs and the covariances between adjacent variables are non-negative, then a reordering of the variables can be found to make the resulting covariance matrix totally positive. The paper also identifies this reordering and provides some necessary and sufficient conditions for the covariance matrix of a multivariate Gaussian distribution to be totally positive.
JOURNAL OF COMPUTATIONAL AND APPLIED MATHEMATICS
(2023)
Article
Engineering, Electrical & Electronic
Ran Li, Jiming Lin, Hongbing Qiu, Junyi Wang
Summary: In recent years, there has been significant interest in distributed estimation of graph Laplacian matrices for smooth graph signals. This paper proposes a distributed graph Laplacian matrix estimation method called the distributed combinatorial graph Laplacian estimation (DCGL), which aims to reduce computational complexity while maintaining estimation accuracy. The method formulates a local parameter estimation problem for each vertex and incorporates Laplacian and structural constraints to resolve the non-convexity between local and global estimates. Experimental results demonstrate the effectiveness of the proposed method in both classical and high-dimensional regimes.
Article
Psychology, Multidisciplinary
Yuki Kobayashi, Akiyoshi Kitaoka
Summary: This study improved the performance of the MIR model by modifying its inference process and priors, resulting in successful predictions of various lightness illusions and phenomena. The study demonstrates the high extensibility and potential of MIR, laying the foundation for further advancements in the field.
FRONTIERS IN PSYCHOLOGY
(2022)
Article
Computer Science, Information Systems
Jose I. de la Rosa, Osvaldo Gutierrez, Jesus Villa-Hernandez, Gamaliel Moreno, Efren Gonzalez, Daniel Alaniz
Summary: In this work, a new robust image segmentation approach is introduced by combining Markov Random Field and non-parametric density estimation strategies within a Bayesian framework. Experimental results show that the proposed method, named MAP entropy estimator (MAPEE), achieves very satisfactory segmentation results when dealing with images degraded with impulsive noise and other non-Gaussian distributions compared to recent robust approaches based on fuzzy c-means segmentation.
MULTIMEDIA TOOLS AND APPLICATIONS
(2021)
Article
Geochemistry & Geophysics
Junxing Yang, Lulu Liu, Qingsong Yan, Fei Deng
Summary: This research proposes an efficient algorithm for generating seamless and large-scale digital orthophoto maps by creating seamline networks from multiple aerial images with overlapping regions. The algorithm consists of two steps: triangulating the overlapping regions and assigning textures to the triangles. Through experimental results, it has been demonstrated that the proposed algorithm can generate high-quality seamline networks efficiently, outperforming existing methods and software.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2022)
Article
Geography, Physical
Yuli Sun, Lin Lei, Xiang Tan, Dongdong Guan, Junzheng Wu, Gangyao Kuang
Summary: This research proposes an unsupervised image regression method based on the inherent structure consistency between heterogeneous images for change detection in multimodal remote sensing images. The proposed method effectively addresses the problem of comparing heterogeneous images and achieves improved detection accuracy compared to state-of-the-art algorithms.
ISPRS JOURNAL OF PHOTOGRAMMETRY AND REMOTE SENSING
(2022)
Article
Geosciences, Multidisciplinary
Fabio Divino, Denekew Bitew Belay, Nico Keilman, Arnoldo Frigessi
Summary: In this study, a spatial approach was used to investigate mortality data referenced over a Lexis structure, decomposing mortality into two interpretable components. Through a hierarchical Bayesian model, the primary structured mortality and secondary additional mortality in Italy and Sweden were estimated, revealing an interesting band of extra mortality in the age interval between 60 and 90 years.
SPATIAL STATISTICS
(2021)
Article
Remote Sensing
Huan Luo, Quan Zheng, Lina Fang, Yingya Guo, Wenzhong Guo, Cheng Wang, Jonathan Li
Summary: This paper proposes a unified neural network for interactive 3D object segmentation from point clouds to reduce manual annotation efforts. By incorporating a boundary energy term in the MRF model, the segmentation performance for accurate object segmentation is improved. Experimental results demonstrate that the proposed method is superior and effective in 3D object segmentation in various point-cloud scenarios.
INTERNATIONAL JOURNAL OF APPLIED EARTH OBSERVATION AND GEOINFORMATION
(2021)
Article
Computer Science, Artificial Intelligence
Xuan Dong, Weixin Li, Xiaojie Wang
Summary: This study proposes a method to address the colorization problem in a multi-lens camera system by combining pyramid processing with CNNs, resulting in a smaller model size and outperforming all state-of-the-art methods in accuracy.
Article
Mathematics, Applied
Juan Baz, Pedro Alonso, Raill Perez-Fernandez
Summary: This paper addresses the matrix construction problem of Gaussian Markov Random Fields with uniform correlation. It provides a characterization of the correlation matrix for a Gaussian Markov Random Field with uniform correlation over a cycle graph, which is circulant and has a sparse inverse matrix. The relationship with the stationary Gaussian Markov Process on the circle is also studied, and two methods for computing the correlation matrix are provided. Asymptotic results for cycle graphs of large order reveal the connection between Gaussian Markov Random Fields with uniform correlation over cycle and path graphs.
LINEAR ALGEBRA AND ITS APPLICATIONS
(2023)
Article
Computer Science, Interdisciplinary Applications
Aryan Eftekhari, Dimosthenis Pasadakis, Matthias Bollhoefer, Simon Scheidegger, Olaf Schenk
Summary: A novel l(1)-regularized maximum likelihood method for efficient large-scale sparse precision matrix estimation utilizing block structures in computations has been proposed. The method demonstrates significant acceleration in computation of covariance matrices, by two to three orders of magnitude, while maintaining modest memory requirements. The applicability of the method for real-world datasets in finance and medicine has been demonstrated through large-scale case studies.
JOURNAL OF COMPUTATIONAL SCIENCE
(2021)
Review
Engineering, Civil
Aditya Pandey, Ashmeet Singh, Paolo Gardoni
Summary: This paper reviews the diagrammatic perturbation theory, a technique in Information Field Theory, for analytically estimating moments of perturbative non-Gaussian distributions. When dealing with physical phenomena, which often exhibit non-Gaussian features, approximation of the underlying distribution and inference of its parameter form are commonly used. More rigorous analysis methods such as Markov Chain Monte Carlo can also be employed, but are computationally expensive.
Article
Computer Science, Artificial Intelligence
Ahmad Khajenezhad, Hatef Madani, Hamid Beigy
Summary: The article proposes two autoencoders for estimating the density of a small set of observations, modifying the masking process of MADE according to the Markov random field (MRF) structure. These modifications help to reduce either the model complexity or the problem complexity.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2021)
Article
Geochemistry & Geophysics
Hossein Aghababaei, Giampaolo Ferraioli
Summary: Despeckling is a necessary task for fully utilizing multichannel SAR images. Various despeckling techniques rely on the estimation of the covariance matrix. This study presents nonreference indices for evaluating polarimetric/interferometric covariance matrices, showing agreement with visual evaluation and common quality measures.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2021)
Article
Geochemistry & Geophysics
Sergio Vitale, Giampaolo Ferraioli, Vito Pascazio
Summary: In the framework of deep learning for synthetic aperture radar (SAR) speckle reduction, the study compared different training approaches to analyze their benefits and drawbacks, with results showing the characteristics of each approach on real SAR images.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2022)
Article
Geochemistry & Geophysics
Hossein Aghababaei, Giampaolo Ferraioli, Sergio Vitale, Roghayeh Zamani, Gilda Schirinzi, Vito Pascazio
Summary: A model-free despeckling framework is proposed in this article, which utilizes similarity distribution and importance to effectively denoise various SAR images while preserving structures and textures.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2022)
Article
Geochemistry & Geophysics
Sergio Vitale, Giampaolo Ferraioli, Vito Pascazio
Summary: This article introduces the application of deep learning techniques in remote sensing, specifically in the despeckling of synthetic aperture radar (SAR) images. It proposes a convolutional neural network (CNN) structure that takes into account both the spatial and statistical properties of SAR images, based on a multi-objective cost function. Experimental results show that the proposed method achieves more accurate despeckling effects.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2021)
Article
Chemistry, Multidisciplinary
Stefano Franceschini, Michele Ambrosanio, Angelo Gifuni, Giuseppe Grassini, Fabio Baselice
Summary: This paper introduces an experimental multiple-input-multiple-output ultrasound tomographic database for testing imaging techniques with various objects of different shapes, sizes, and materials.
APPLIED SCIENCES-BASEL
(2022)
Article
Chemistry, Analytical
Michele Ambrosanio, Stefano Franceschini, Maria Maddalena Autorino, Fabio Baselice, Vito Pascazio
Summary: The advancement of new technologies in biomedical imaging is crucial for the research community, and ultrasound tomography has proven to be a safe diagnostic tool beneficial for breast cancer imaging with its operator-independent feature.
Article
Biotechnology & Applied Microbiology
Stefano Franceschini, Michele Ambrosanio, Vito Pascazio, Fabio Baselice
Summary: This paper presents a novel ultrasound system for person identification using hand gestures. The system works by measuring the ultrasonic pressure waves scattered by the subject's hand and analyzing its Doppler information. The acquired signal undergoes several transformations to obtain time/frequency representations, and a deep learning detector is implemented. The system is cheap, reliable, contactless, and can be easily integrated with other personal identification approaches, allowing for different levels of security. Experimental tests on 25 volunteers showed promising results, demonstrating the potential of the system.
BIOENGINEERING-BASEL
(2023)
Article
Biotechnology & Applied Microbiology
Michele Ambrosanio, Stefano Franceschini, Vito Pascazio, Fabio Baselice
Summary: This paper proposes an artificial neural network approach for effective and real-time quantitative microwave breast imaging, and provides numerical analyses for optimizing network architecture and improving recovery performance and processing time.
BIOENGINEERING-BASEL
(2022)
Article
Medicine, General & Internal
Stefano Franceschini, Maria Maddalena Autorino, Michele Ambrosanio, Vito Pascazio, Fabio Baselice
Summary: In this paper, a deep learning technique is proposed for tumor detection in a microwave tomography framework. Micrwave tomography has gained attention for its ability to reconstruct internal breast tissue properties using nonionizing radiations. The proposed approach utilizes deep learning to identify the presence of tumors based on tomographic measures, showing promising performance for small tumor masses. This method can be used for early diagnosis where the mass being detected may be particularly small.
Article
Biotechnology & Applied Microbiology
Marijn Borghouts, Michele Ambrosanio, Stefano Franceschini, Maria Maddalena Autorino, Vito Pascazio, Fabio Baselice
Summary: This study demonstrates the potential of using a convolutional neural network to convert scattering matrices into tumor probability maps, leading to significant advancements in breast cancer screening.
BIOENGINEERING-BASEL
(2023)
Article
Geochemistry & Geophysics
Hossein Aghababaei, Giampaolo Ferraioli, Alfred Stein, Sergio Vitale
Summary: Synthetic Aperture Radar (SAR) systems can have different polarimetric modalities, but most spaceborne SAR systems use dual-polarimetric data to capture more information about the Earth's surface and cover a wider area. This article proposes a new framework that uses deep learning to reconstruct fully polarimetric information from typical dual-pol data without relying on model assumptions. Experimental results show that the proposed framework outperforms traditional reconstruction methods and the reconstructed pseudo-fully polarimetric data closely matches actual fully polarimetric images acquired by radar systems, confirming the reliability and effectiveness of the proposed method.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Geochemistry & Geophysics
Wenyu Yang, Sergio Vitale, Hossein Aghababaei, Giampaolo Ferraioli, Vito Pascazio, Gilda Schirinzi
Summary: This article presents a deep learning approach called TSNN for reconstructing forest and ground height using multipolarimetric multibaseline SAR data and LiDAR-based data. TSNN is trained with covariance matrix elements as input vectors and quantized LiDAR data as the reference, and it formulates the height reconstruction as a classification problem. Experimental results show that TSNN achieves high spatial resolution and vertical accuracy in height measurement, outperforming other TomoSAR methods.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Geochemistry & Geophysics
Wenyu Yang, Sergio Vitale, Hossein Aghababaei, Giampaolo Ferraioli, Vito Pascazio, Gilda Schirinzi
Summary: Forest characterization and monitoring are important for tracking climate change and biodiversity. TomoSAR is a powerful tool for reconstructing 3-D structures. This study extends the TomoSAR neural network to retrieve forest height and underlying topography, without requiring fully polarimetric data.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2023)
Article
Geochemistry & Geophysics
Sergio Vitale, Giampaolo Ferraioli, Alejandro C. Frery, Vito Pascazio, Dong-Xiao Yue, Feng Xu
Summary: This paper proposes a deep-learning based method for despeckling synthetic aperture radar (SAR) images. By using a multicategory generalized Gaussian coherent SAR simulator to construct diversified training datasets and designing an effective multiobjective cost function, the method demonstrates superior performance on multiple SAR datasets.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Geochemistry & Geophysics
Sergio Vitale, Giampaolo Ferraioli, Vito Pascazio, Gilda Schirinzi
Summary: This article proposes a new interferometric phase denoising algorithm, InSAR-MONet, which removes noise and preserves important phase details by defining a multiobjective cost function. The algorithm is evaluated on simulated and real datasets and compared with state-of-the-art interferometric denoising algorithms.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2022)