Incorporating reference guided priors into calibrationless parallel imaging reconstruction

Joint transmitter and receiver beamforming algorithms for uplink multiuser multi-input multi-output (MIMO) communication systems are proposed in this paper based on maximizing an utility function, which is defined based on the difference between desired signal power and interference plus noise power. Two iterative beamforming algorithms are developed by maximizing the utility function of each user via parallel and serial methods. In parallel method, the transmitter and receiver beamforming vectors of each user are iteratively updated by maximizing its utility function. However in serial method, at first step, the transmitter beamforming vectors of all users are iteratively updated by maximizing the utility function of each user and then receiver beamforming vectors of users are calculated. By computer simulations, the performances of the proposed beamforming algorithms are evaluated and also compared with the performance of the duality method (DM) based beamforming algorithm. With respect to bit error rate (BER) and convergence rate, the results show that the new proposed beamforming algorithms have superior performances in comparison with the DM based beamforming algorithm. Also, for some cases, the serial method achieves better performances in comparison with the parallel method.

Thick cloud contamination in remote sensing (RS) images significantly hinders their utility in subsequent applications. Traditional cloud removal methods predominantly focus on the design of the regularization term, while neglecting the design of the fidelity term. Recently, the proposal of gradient-domain fidelity has highlighted the significance of the fidelity term design, dedicated to maintaining textures in the gradient domain. However, the handcrafted gradient-domain fidelity still has limitations in capturing invariant and delicate features behind multitemporal RS (MTRS) images, leading to unsatisfactory detail preservation. To address the problem, we suggest a deep domain fidelity to capture deep features by leveraging a pretrained deep network, which matches the intrinsic deep feature between the original images and reference images, instead of matching the shallow features in gradient domain fidelity. Empowered with the deep domain fidelity, we propose a thick cloud removal model (called DFTR) for MTRS images, organically integrating the deep domain fidelity term with a low-rank (LR) regularization term (i.e., tensor ring (TR) decomposition), offering fine detail preservation. Extensive simulated and real experiments on MTRS images demonstrate that the proposed method outperforms the compared methods, including the origin domain-based and the gradient domain-based methods, in thick cloud removal, especially for detail preservation.

High-quality reference images can be seen as a baseline for many image reconstruction techniques. The reconstruction performance gets improved as shown in the Reference Image Constrained Regularization by Denoising. But, low-quality reference images lead to disappointing results, even underperforming methods that use no reference at all. To address this limitation, a novel framework termed Adaptive Reference Constrained Regularization by Denoising is proposed in this paper. This new method mitigates the influence of low-quality reference images by adaptively leveraging the reconstruction results obtained from the existing algorithm to form a more reliable constraint. Furthermore, an enhanced version is introduced, equipped with a self-tuning parameter that progressively diminishes the weight of a poor reference throughout the optimization iterations. Extensive experiments on single-image super-resolution and MRI reconstruction demonstrate that the proposed approach significantly outperforms state-of-the-art methods, including Regularization by Denoising and Reference Image Constrained Regularization by Denoising, in both quantitative metrics and visual fidelity.

Purpose: The aim of this work is to investigate the feasibility of accelerating phase contrast MR angiography by the fast imaging method of Simplified Skipped Phase Encoding and Edge Deghosting with Array Coil Enhancement (S‐SPEED‐ACE). Methods: The parallel imaging method of Skipped Phase Encoding and Edge Deghosting with Array Coil Enhancement (SPEED‐ACE) has been simplified with enhanced acceleration for imaging sparse objects and is termed S‐SPEED‐ACE. As demonstrated previously with a computer simulated study, S‐SPEED‐ACE partially samples k‐space with multiple coils in parallel and yields a deghosted image based on a least‐square‐error solution. Without differential filtering and full central k‐space sampling, S‐SPEED‐ACE uses more straightforward reconstruction and achieves further scan time reduction as compared with SPEED‐ACE. In this work, S‐SPEED‐ACE is further developed to accelerate real human 3D phase‐contrast MR angiography. The proposed method was tested with a 3D head PC‐MRA scan, which was acquired using a clinical 1.5T scanner on a healthy volunteer Results: Images are reconstructed by S‐SPEED‐ACE to achieve an undersampling factor of up to 8.3 with four receiver coils. The reconstructed images generally have comparable quality as that of the reference imagesreconstructed from full k‐space data. Maximum‐intensity‐projection (MIP) images generated from the reconstructed images also demonstrated to be consistent as those from the reference images. Conclusions: By taking advantage of signal sparsity naturally existing in the data, SPEED‐ACE was simplified and its efficiency was improved. In this work, the previously proposed parallel imaging method named S‐SPEED‐ACE has been further developed to accelerate a 3D head PC‐MRA scan.

Image quality assessment method is a methodology that measures the difference of quality between the reference image and its distorted one. In this paper, we propose a novel reduced-reference (RR) quality assessment method for JPEG-2000 compressed images, which exploits the statistical characteristics of context information extracted through partial entropy decoding or decoding. These statistical features obtained in the process of JPEG-2000 encoding are transmitted to the receiver as side information and used to estimate the quality of images transmitted over various noisy channels at the decompression side. In the framework of JPEG-2000, the context of a current coefficient is determined depending on the pattern of the significance and/or the sign of its neighbors in three bit-plane coding passes and four coding modes. As the context information represents the local property of images, it can efficiently describe textured pattern and edge orientation. The quality of transmitted images is measured by the difference of entropy of context information between received and original images. Moreover, the proposed quality assessment method can directly process the images in the JPEG-2000 compressed domain without full decompression. Therefore, our proposed can accelerate the work of assessing image quality. Through simulations, we demonstrate that our method achieves fairly good performance in terms of the quality measurement accuracy as well as the computational complexity.

We conducted a six-month investigation of the design, analysis, and software implementation of a class of singularity-insensitive, scalable, parallel nonlinear iterative methods for the numerical solution of nonlinear partial differential equations. The solutions of nonlinear PDEs are often nonsmooth and have local singularities, such as sharp fronts. Traditional nonlinear iterative methods, such as Newton-like methods, are capable of reducing the global smooth nonlinearities at a nearly quadratic convergence rate but may become very slow once the local singularities appear somewhere in the computational domain. Even with global strategies such as line search or trust region the methods often stagnate at local minima of {parallel}F{parallel}, especially for problems with unbalanced nonlinearities, because the methods do not have built-in machinery to deal with the unbalanced nonlinearities. To find the same solution u* of F(u) = 0, we solve, instead, an equivalent nonlinearly preconditioned system G(F(u*)) = 0 whose nonlinearities are more balanced. In this project, we proposed and studied a nonlinear additive Schwarz based parallel nonlinear preconditioner and showed numerically that the new method converges well even for some difficult problems, such as high Reynolds number flows, when a traditional inexact Newton method fails.

Transformer-based models like large language models (LLMs) have attracted significant attention in recent years due to their superior performance. A long sequence of input tokens is essential for industrial LLMs to provide better user services. However, memory consumption increases quadratically with the increase of sequence length, posing challenges for scaling up long-sequence training. Current parallelism methods produce duplicated tensors during execution, leaving space for improving memory efficiency. Additionally, tensor parallelism (TP) cannot achieve effective overlap between computation and communication. To solve these weaknesses, we propose a general parallelism method called memory-efficient tensor parallelism (METP), designed for the computation of two consecutive matrix multiplications and a possible function between them (O = f(AB)C), which is the kernel computation component in Transformer training. METP distributes subtasks of computing O to multiple devices and uses send/recv instead of collective communication to exchange submatrices for finishing the computation, avoiding producing duplicated tensors. We also apply the double buffering technique to achieve better overlap between computation and communication. We present the theoretical condition of full overlap to help instruct the long-sequence training of Transformers. Suppose the parallel degree is p; through theoretical analysis, we prove that METP provides O(1/p3) memory overhead when not using FlashAttention to compute attention and could save at least 41.7% memory compared to TP when using FlashAttention to compute multi-head self-attention. Our experimental results demonstrate that METP can increase the sequence length by 2.38–2.99 times compared to other methods when using eight A100 graphics processing units (GPUs).

In this paper, an approach to medical image registration using Powell's algorithm with a basic concept from mutual information, & entropy, as a new matching criterion is presented. This approach uses histogram equalised reference image and target image. Volume control points of these enhanced images determine the quality of image registration. Based on these volume control points, features like location, edge pixel intensity strength & orientation are considered to compute a joint probability distribution of corresponding edge points from reference and target images. Then mutual information based on this function is minimised to find the best alignment parameters and the translation parameters are calculated using Powell's algorithm and matched to perform image registration. The proposed registration algorithm is faster, robust & proved to be more efficient than the ACO approach quantitatively. Simulations for Powell's algorithm using enhanced images of the same size but with different angles are shown here.

In this paper, we present a robust exposure fusion algorithm to tackle the problems of motion removal and detail preserving in dynamic scenes. With one exposure as reference, the motion appeared in the exposure stack can be detected by comparing the structural consistency, which is extracted by measuring the degree of linear correlation between the patches of the reference image and the other source images. Then, a stack of latent images with consistent contents can be synthesized after motion removal. For detail preserving, a contrast criterion is introduced to measure the exposedness and generate visibility maps of each latent image. Guided by the visibility maps, a tonemapped-like HDR image which is ghost-free and with all details preserved could be produced by seamlessly merging the latent images. Exposure fusion tests on various dynamic scenes demonstrate the superiority of the proposed method over existing state-of-the-art approaches.

Precise delineation of tumor tissues during surgery is hindered by the strong scattering and absorption of photons in biological tissues, which obscure molecular imaging signals and hinder intraoperative navigation. Raman bioimaging using surface-enhanced Raman scattering (SERS) nanoprobes, offers high molecular specificity and sensitivity but still suffers from severe degradation when the signals transmitted through scattering media. Here, we present a deep learning-assisted approach for reconstructing Raman speckle images obscured by biological scattering. We synthesized gap-enhanced resonance Raman tags (GERRTs) and used them to draw Raman patterns on paper. These patterns were tested to build the dataset of image pairs, including reference Raman images that collected directly from the patterns and speckle Raman images after signal transmitting through 4-mm porcine tissues. To recover these images, we employed a U-Net architecture featuring symmetric encoder-decoder pathways with skip connections, which effectively preserves high-resolution spatial details lost in scattering tissue. This model achieved superior reconstruction performance, with an average structural similarity index measure (SSIM) of 0.539 between reconstructed images and reference images, and a peak signal-to-noise ratio (PSNR) of 19.04 dB. This work provides a proof-of-concept framework for deep learning-assisted Raman imaging through scattering tissues and holds promise for advancing noninvasive optical molecular imaging and facilitating precise surgical navigation in oncology.

The increasing reliance on deep learning models in video understanding necessitates transparent and interpretable decision-making, with video counterfactual explanations (CEs) offering a critical avenue to understand model behavior. However, generating effective video CEs remains challenging due to video data's high dimensionality, temporal coherence demands, and the need to balance precise alterations with visual realism. Existing Latent Diffusion Model (LDM)-based CE methods often struggle with generation efficiency, exact target adherence, and the suppression of subtle visual artifacts. To address these limitations, we propose an Adaptive-Guided Latent Diffusion Model for Counterfactual Explanations (AG-LDM-CE). Our framework introduces an Adaptive Gradient Guidance (AGG) mechanism that dynamically adjusts guidance strength based on proximity to the target prediction, optimizing efficiency and balancing target adherence with visual fidelity. Complementing this, a novel Multi-Scale Perceptual Refinement (MSPR) module leverages multi-level VAE features to intelligently suppress artifacts and ensure that counterfactual changes are accurately localized to causally relevant regions. Extensive evaluations across diverse video regression (EchoNet-Dynamic) and classification (FERV39K, Something-Something V2) tasks demonstrate AG-LDM-CE's superior performance. Our method significantly improves generation efficiency and explanation quality, achieving strong target adherence and perceptual realism. Ablation studies and human evaluations further validate the significant contributions of AGG and MSPR, confirming that AG-LDM-CE generates more efficient, accurate, perceptually realistic, and precisely localized video counterfactual explanations.

Diffusion-based text-to-image generation models like GLIDE and DALLE-2 have gained wide success recently for their superior performance in turning complex text inputs into images of high quality and wide diversity. In particular, they are proven to be very powerful in creating graphic arts of various formats and styles. Although current models supported specifying style formats like oil painting or pencil drawing, fine-grained style features like color distributions and brush strokes are hard to specify as they are randomly picked from a conditional distribution based on the given text input. Here we propose a novel style guidance method to support generating images using arbitrary style guided by a reference image. The generation method does not require a separate style transfer model to generate desired styles while maintaining image quality in generated content as controlled by the text input. Additionally, the guidance method can be applied without a style reference, denoted as self style guidance, to generate images of more diverse styles. Comprehensive experiments prove that the proposed method remains robust and effective in a wide range of conditions, including diverse graphic art forms, image content types and diffusion models.

In order to solve the problems of scale difference, inconspicuous spatial feature expression, redundancy, and overlapping of spectral feature information in hyperspectral and multispectral image fusion, a fusion algorithm based on multi-scale space-spectrum hybrid attention network is designed in this paper. The algorithm enhances the capability of shallow feature extraction through the multi-scale mechanism, uses the space-spectrum hybrid attention mechanism to mine the correlation between deep space and spectrum, designs a cross-spectrum fusion module to realize the effective reconstruction of spatial and spectral information, and uses the loss function to constrain the difference between the fused image and the reference image in terms of color, detail edge, and spatial structure. Experiments were conducted on PaviaUniversity, IndianPines, and hyperspectral image of Natural Scenes2004 datasets, and compared to algorithms such as ResTFNet, spatial spectral reconstruction network, MoGDCN, DBSR, and Fusformer. The proposed algorithm, MSSSHANet, performs better in spectral curve smoothness, spectral difference value, root mean squared error, peak signal to noise ratio, Erreur Relative Globale Adimensionnelle de Synthèse, and Spectral Angle Mapper value, which plays an active role in improving fusion quality. However, the generalization ability of the proposed algorithm in special remote sensing tasks needs to be verified, and future research will focus on data preprocessing optimization and the development of time-spatial-spectral integration fusion technology.

Remote sensing-based forest tree species classification plays a crucial role in forest resource management and ecological monitoring. In recent years, hyperspectral reconstruction technology has emerged as an effective approach to enhance the spectral resolution of multispectral imagery. By leveraging the advantages of both multispectral and hyperspectral data, this technique offers the potential to improve tree species classification accuracy. However, the application of hyperspectral reconstruction methods in forestry remains limited, with existing approaches often suffering from insufficient spectral information utilization and a strong reliance on large-scale datasets. To address the specific demands of hyperspectral reconstruction in forestry, this study proposes HSCNN-Unet, a model that employs U-Net as its overall architecture and integrates HSCNN-D for spectral reconstruction. Additionally, a Spectral Attention mechanism and a feature fusion module are incorporated to enhance spectral feature extraction and improve data reconstruction capability. We evaluated the proposed model on multiple sample sets and in comparison, with various baseline models. Experimental results show that HSCNN-Unet demonstrates superior spectral reconstruction performance on smaller-sized samples compared to all baseline models (MPSNR values of 51.9587 and 49.1999, MSSIM of 0.9954, and SAM values of 0.0141 and 0.0213), while still maintaining strong reconstruction performance on larger-sized samples. In addition, the effectiveness of the reconstructed hyperspectral data in classification tasks was also verified. Furthermore, the model was applied to the spectral reconstruction of Sentinel-2 imagery and tested in combination with dominant tree species classification in the study area. The experimental results show that compared to the classification performance of the original multispectral data, the reconstructed hyperspectral data achieved improvements of 6.72%, 6.71%, and 8.88% in Weighted F1, Overall Accuracy, and Kappa, respectively. Other metrics also improved by 8.95%–9.20%. These results demonstrate the excellent reconstruction performance and practical classification utility of HSCNN-Unet, confirming its real-world application value in forest resource monitoring.

Face hallucination, which refers to the restoration of single or multiple low-resolution face images into clear high-resolution one, is a challenging research issue. Most existing methods use local or global patches for image representation and achieve good performance. However, they ignore that the local patch limits the area of the receptive field and the priori information used in reconstruction is limited. And the global patch expands the receptive field but introduces irrelevant information to degrade the reconstruction performance. In order to improve the performance of reconstruction, we propose a nonlinear contextual face hallucination method. First, contextual information can effectively improve the receptive field area to make full use of priori information. Then, the nonlinear model can make the proposed model more suitable for practical application and make the correlation of data in kernel space more compact. Finally, combining contextual and residual learning can improve the stability of the solution of the super-resolution model and the accuracy of reconstruction performance. The experimental results show that the proposed face hallucination method has superior performance than the state-of-the-art method.