Bio: Yun-Heng Wang is an academic researcher from Harbin Institute of Technology. The author has contributed to research in topic(s): Sparse approximation & Kernel embedding of distributions. The author has an hindex of 3, co-authored 3 publication(s) receiving 87 citation(s).
01 Jan 2014-Measurement
TL;DR: A novel dictionary training method for sparse reconstruction for enhancing the similarity of sparse representations between the low resolution and high resolution MRI block pairs through simultaneous training two dictionaries.
Abstract: Magnetic Resonance Imaging (MRI) data collection is influenced by SNR, hardware, image time, and other factors. The super-resolution analysis is a critical way to improve the imaging quality. This work presents a framework of super-resolution MRI via sparse reconstruction, and this method is promising to solve the data collection limitations. A novel dictionary training method for sparse reconstruction for enhancing the similarity of sparse representations between the low resolution and high resolution MRI block pairs through simultaneous training two dictionaries. Low resolution MRI blocks generate the high resolution MRI blocks with proposed sparse representation (SR) coefficients. Comprehensive evaluations are implemented to test the feasibility and performance of the SR–MRI method on the real database.
01 Feb 2014-Information Sciences
TL;DR: In this paper, a uniform framework for kernel self-optimization with the ability to adjust the data structure is presented, where the data-dependent kernel is extended and applied to kernel learning, and optimization equations with two criteria for measuring data discrimination are used to solve the optimal parameter values.
Abstract: Kernel learning is becoming an important research topic in the area of machine learning, and it has wide applications in pattern recognition, computer vision, image and signal processing. Kernel learning provides a promising solution to nonlinear problems, including nonlinear feature extraction, classification and clustering. However, in kernel-based systems, the problem of the kernel function and its parameters remains to be solved. Methods of choosing parameters from a discrete set of values have been presented in previous studies, but these methods do not change the data distribution structure in the kernel-based mapping space. Accordingly, performance is not improved because the current kernel optimization does not change the data distribution. Based on this problem, this paper presents a uniform framework for kernel self-optimization with the ability to adjust the data structure. The data-dependent kernel is extended and applied to kernel learning, and optimization equations with two criteria for measuring data discrimination are used to solve the optimal parameter values. Some experiments are performed to evaluate the performance in popular kernel learning methods, including kernel principal components analysis (KPCA), kernel discriminant analysis (KDA) and kernel locality-preserving projection (KLPP). These evaluations show that the framework of kernel self-optimization is feasible for enhancing kernel-based learning methods.
••18 Jul 2012
TL;DR: A novel method for Super-Resolution Medical image based sparse representation with two coupled dictionaries to solve the problem of MR image resolution owing to the limitations of hardware and acquisitions is proposed.
Abstract: In this paper, we propose a novel method for Super-Resolution Medical image based sparse representation, with the aim to solve the problem of MR image resolution owing to the limitations of hardware and acquisitions. With two coupled dictionaries the sparse representation of a low resolution medical image blocks is used to generate a high resolution. Some evaluations are implemented to compare with previous method, and the proposed algorithm has its advantage on super-resolution.
01 Nov 2016-Signal Processing
TL;DR: This paper aims to provide a review of SR from the perspective of techniques and applications, and especially the main contributions in recent years, and discusses the current obstacles for future research.
Abstract: Super-resolution (SR) technique reconstructs a higher-resolution image or sequence from the observed LR images As SR has been developed for more than three decades, both multi-frame and single-frame SR have significant applications in our daily life This paper aims to provide a review of SR from the perspective of techniques and applications, and especially the main contributions in recent years Regularized SR methods are most commonly employed in the last decade Technical details are discussed in this article, including reconstruction models, parameter selection methods, optimization algorithms and acceleration strategies Moreover, an exhaustive summary of the current applications using SR techniques has been presented Lastly, the article discusses the current obstacles for future research
••17 Jun 2014
TL;DR: The working principles, applications, merits, and demerits of these methods has been discussed in detail along with their other technical issues followed by present status and future trends.
Abstract: Under the alternating electrical excitation, biological tissues produce a complex electrical impedance which depends on tissue composition, structures, health status, and applied signal frequency, and hence the bioelectrical impedance methods can be utilized for noninvasive tissue characterization. As the impedance responses of these tissue parameters vary with frequencies of the applied signal, the impedance analysis conducted over a wide frequency band provides more information about the tissue interiors which help us to better understand the biological tissues anatomy, physiology, and pathology. Over past few decades, a number of impedance based noninvasive tissue characterization techniques such as bioelectrical impedance analysis (BIA), electrical impedance spectroscopy (EIS), electrical impedance plethysmography (IPG), impedance cardiography (ICG), and electrical impedance tomography (EIT) have been proposed and a lot of research works have been conducted on these methods for noninvasive tissue characterization and disease diagnosis. In this paper BIA, EIS, IPG, ICG, and EIT techniques and their applications in different fields have been reviewed and technical perspective of these impedance methods has been presented. The working principles, applications, merits, and demerits of these methods has been discussed in detail along with their other technical issues followed by present status and future trends.
TL;DR: To develop a super‐resolution technique using convolutional neural networks for generating thin‐slice knee MR images from thicker input slices, and compare this method with alternative through‐plane interpolation methods.
Abstract: PURPOSE To develop a super-resolution technique using convolutional neural networks for generating thin-slice knee MR images from thicker input slices, and compare this method with alternative through-plane interpolation methods. METHODS We implemented a 3D convolutional neural network entitled DeepResolve to learn residual-based transformations between high-resolution thin-slice images and lower-resolution thick-slice images at the same center locations. DeepResolve was trained using 124 double echo in steady-state (DESS) data sets with 0.7-mm slice thickness and tested on 17 patients. Ground-truth images were compared with DeepResolve, clinically used tricubic interpolation, and Fourier interpolation methods, along with state-of-the-art single-image sparse-coding super-resolution. Comparisons were performed using structural similarity, peak SNR, and RMS error image quality metrics for a multitude of thin-slice downsampling factors. Two musculoskeletal radiologists ranked the 3 data sets and reviewed the diagnostic quality of the DeepResolve, tricubic interpolation, and ground-truth images for sharpness, contrast, artifacts, SNR, and overall diagnostic quality. Mann-Whitney U tests evaluated differences among the quantitative image metrics, reader scores, and rankings. Cohen's Kappa (κ) evaluated interreader reliability. RESULTS DeepResolve had significantly better structural similarity, peak SNR, and RMS error than tricubic interpolation, Fourier interpolation, and sparse-coding super-resolution for all downsampling factors (p < .05, except 4 × and 8 × sparse-coding super-resolution downsampling factors). In the reader study, DeepResolve significantly outperformed (p < .01) tricubic interpolation in all image quality categories and overall image ranking. Both readers had substantial scoring agreement (κ = 0.73). CONCLUSION DeepResolve was capable of resolving high-resolution thin-slice knee MRI from lower-resolution thicker slices, achieving superior quantitative and qualitative diagnostic performance to both conventionally used and state-of-the-art methods.
01 Jan 2016-Medical Image Analysis
TL;DR: A graph-based redundant wavelet transform is introduced to sparsely represent magnetic resonance images in iterative image reconstructions and outperforms several state-of-the-art reconstruction methods in removing artifacts and achieves fewer reconstruction errors on the tested datasets.
Abstract: Compressed sensing magnetic resonance imaging has shown great capacity for accelerating magnetic resonance imaging if an image can be sparsely represented. How the image is sparsified seriously affects its reconstruction quality. In the present study, a graph-based redundant wavelet transform is introduced to sparsely represent magnetic resonance images in iterative image reconstructions. With this transform, image patches is viewed as vertices and their differences as edges, and the shortest path on the graph minimizes the total difference of all image patches. Using the l1 norm regularized formulation of the problem solved by an alternating-direction minimization with continuation algorithm, the experimental results demonstrate that the proposed method outperforms several state-of-the-art reconstruction methods in removing artifacts and achieves fewer reconstruction errors on the tested datasets.
TL;DR: This work proposes a novel residual learning-based SR algorithm for MRI, which combines both multi-scale GRL and shallow network block-based local residual learning (LRL), which works effectively in capturing high-frequency details by learning local residuals.
Abstract: Spatial resolution is one of the key parameters of magnetic resonance imaging (MRI). The image super-resolution (SR) technique offers an alternative approach to improve the spatial resolution of MRI due to its simplicity. Convolutional neural networks (CNN)-based SR algorithms have achieved state-of-the-art performance, in which the global residual learning (GRL) strategy is now commonly used due to its effectiveness for learning image details for SR. However, the partial loss of image details usually happens in a very deep network due to the degradation problem. In this work, we propose a novel residual learning-based SR algorithm for MRI, which combines both multi-scale GRL and shallow network block-based local residual learning (LRL). The proposed LRL module works effectively in capturing high-frequency details by learning local residuals. One simulated MRI dataset and two real MRI datasets have been used to evaluate our algorithm. The experimental results show that the proposed SR algorithm achieves superior performance to all of the other compared CNN-based SR algorithms in this work.