[02618] Recent Developments in Hyperspectral and Multispectral Imaging
Session Time & Room : 5C (Aug.25, 13:20-15:00) @E605
Type : Proposal of Minisymposium
Abstract : Hyperspectral and multispectral images contain plenties of spatial–spectral information, which brings a substantial opportunity to explore and understand their features. This Mini-symposium provides a platform for researchers to share innovative ideas about algorithm developments associated with this type of imaging. Attendees can expect to learn about the latest techniques from model-based and data-driven perspectives and their applications in remote sensing, computational imaging, and medical imaging.
[03524] Multi-Dimensional Signal Alignment using Local All-Pass Filters
Format : Talk at Waseda University
Author(s) :
Christopher Gilliam (University of Birmingham)
Abstract : The estimation of a geometric transformation that aligns two or more signals is a problem that has many applications. The problem occurs when signals are either recorded from two or more spatially separated sensors or when a single sensor is recording a time-varying scene, e.g., image registration, motion correction in medical imaging and time-varying delay estimation. In this talk we estimate the transformation by approximating it using a set of local all-pass (LAP) filters.
[03596] Noise reduction in X-ray microspectroscopy
Format : Online Talk on Zoom
Author(s) :
Jizhou Li (City University of Hong Kong)
Abstract : Investigating nanoscale morphological and chemical phase transformations is essential for a wide range of scientific and industrial applications across various disciplines. The emerging TXM-XANES imaging technique combines the strengths of full-field transmission X-ray microscopy (TXM) and X-ray absorption near edge structure (XANES) to generate chemical maps by capturing a series of multi-energy X-ray microscopy images and fitting them accordingly. However, its effectiveness is hindered by low signal-to-noise ratios due to system errors and insufficient exposure illuminations during rapid data acquisition. In this study, we present a straightforward and robust denoising method that capitalizes on the inherent properties and subspace modeling of TXM-XANES imaging data to significantly improve image quality, paving the way for fast and highly sensitive chemical imaging. Comprehensive experiments using both synthetic and real datasets showcase the remarkable performance of our proposed approach.
[03540] Remote Sensing Image Reconstruction from the Subspace Perspective
Format : Talk at Waseda University
Author(s) :
Jie Lin (University of Electronic Science and Technology of China)
Abstract : Remote sensing images cover abundant spatial-spectral information while the images usually suffer from different degradations during the imaging and transmission. Image reconstruction is a fundamental step for subsequent applications. With improved imaging accuracy, the larger sizes of acquired images bring a greatly increased computation burden in reconstruction. In this talk, I will present matrix and tensor subspace-based methods for remote sensing image reconstruction, which enjoy satisfactory effects and lower computational complexity.
Lina Zhuang (Key Laboratory of Computational Optical Imaging Technology, Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing)
Lianru Gao (Key Laboratory of Computational Optical Imaging Technology, Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing)
Bing Zhang (Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China, and also with the University of Chinese Academy of Sciences, Beijing)
Abstract : Deep-learning-based denoising methods for hyperspectral images have been comprehensively studied and achieved impressive performance. Compared with deep-learning-based methods, the nonlocal similarity-based methods are more suitable for images containing edges or regular textures. We propose a powerful denoising method, termed non-local 3-D convolutional neural network, combining traditional machine learning and deep learning techniques. The numerical and graphical denoising results of the simulated and real data show that the proposed method is superior to the state-of-the-art methods.
