Search

Positron Emission Tomography (PET) image reconstruction remains a pivotal area in PET technology, critically influencing clinical diagnostic outcomes. Addressing the need for enhanced image quality, this study introduces a novel algorithm for PET image reconstruction. This algorithm integrates a penalty mechanism, morphological filtering, and the Maximum Likelihood Expectation Maximization (MLEM) algorithm, aiming to improve reconstructed image quality significantly. The operational process of the algorithm within each iteration encompasses two primary phases. Initially, image reconstruction is accomplished via the MLEM algorithm, followed by the application of a morphological filter to attenuate noise in the reconstructed image. Simulation experiments demonstrate that this algorithm effectively mitigates noise while preserving crucial details, such as image edges. Notably, this method presents the dual benefits of straightforward parameter configuration and ease of implementation. The results indicate a substantial enhancement in noise suppression and fine structure preservation in the reconstructed images, marking a significant advancement in PET image reconstruction techniques. [ABSTRACT FROM AUTHOR]

To precisely estimate blur kernels (BKs) and obtain high-quality latent images (LIs), we present an efficient blind image deblurring (BID) algorithm via a gradient prior and an enhanced sparse prior. The proposed algorithm decomposes BID into two phases: multiscale blur estimation (MSBE) and image deconvolution (ID). The former focuses on precise BKs, whereas the latter focuses on high-quality LIs. In the MSBE phase, significant image gradients and an enhanced sparse prior are used to estimate BKs. To this end, we build an MSBE model using a hyper-Laplacian gradient (HLG) prior and sparsity-inducing l2,1-regularization. l2,1-regularization imposes enhanced sparsity on the BKs to be estimated, and the HLG prior characterizes the true distribution of image gradients. We use a method derived from half-quadratic splitting (HQS) to handle the built MSBE model by decoupling it into simpler subproblems. Subsequently, these subproblems are solved by efficient and concise methods. In the MSBE phase, we also adopt a multiscale process to avoid local optima of the estimated BKs and adaptively update a penalty parameter to accelerate the iteration. In the ID phase, SotA restoration algorithms are employed by the proposed BID algorithm to obtain competitive final LIs. The experiments are conducted on several baseline datasets to evaluate our BID algorithm. Under the evaluation criteria of success-rate (SR), SSDE, PSNR, SSIM, vision effect, and running time (RT), our study shows apparent advantages compared with some SotA BID algorithms. These advantages include but are not limited to reducing the RT; obtaining higher PSNRs, SSIMs, and SRs; decreasing SSDEs; and obtaining high-quality LIs without artifacts. [ABSTRACT FROM AUTHOR]

Blind image deconvolution has garnered significant attention in the realm of image processing. Among existing blind deconvolution models, the adaptive weighted total variationbased model can better recover local image features. However, this model may give rise to undesirable staircase effects. In contrast, when considering non-blind deconvolution, a model that uses adaptive Euler's elastica (AEE) regularization can reduce the appearance of staircase effects and better maintain image local features. Inspired by the successful application of the AEE regularization in non-blind deconvolution, we propose an AEE model for blind image deconvolution. This novel model can preserve image features in smooth regions and restore local characteristics of the image. In addition, an efficient alternating direction method of multipliers is designed to address this nonconvex model. Experimental results illustrate the superiority of the proposed method compared with other related blind deconvolution methods. [ABSTRACT FROM AUTHOR]

In compressive sensing theory, the measurement matrix plays a crucial role in compressive observation of sparse signals. The bipolar Toeplitz measurement matrix constructed based on chaotic map has advantages such as generating fewer free elements and supporting fast algorithms, making it widely used. While optimizing the measurement matrix can effectively improve its compressive sensing reconstruction performance, existing optimization algorithms are not suitable for the bipolar Toeplitz measurement matrix due to its structural and bipolar properties. To address this issue, this paper proposes an optimization method for the bipolar Toeplitz measurement matrix based on cosine-exponential (CE) chaotic map sequences and an improved Abolghasemi algorithm. Using an enhanced CE chaotic map to generate chaotic sequences with greater chaos and randomness, we construct the measurement matrix and optimize it using the structure matrix and the improved Abolghasemi algorithm, which preserves the matrix's bipolarity without altering its structure. We also introduce constraints on the generated sequence values during the optimization process. Through simulation experiments, the effectiveness of our optimization algorithm is verified, as the optimized bipolar Toeplitz measurement matrix significantly reduces reconstruction error and improves reconstruction probability. [ABSTRACT FROM AUTHOR]

The Graph Coloring Problem is a well-known NP-hard problem. Over the years, numerous scholars have been pursuing efficient algorithms to obtain high-quality solutions. Nested Rollout Policy Adaptation (NRPA) is a Monte Carlo Tree Search algorithm for single-player games, and it has been proven effective and good in combinatorial optimization problems. In this paper, we use for the first time the NRPA algorithm combined with the destruction and reconstruction ideas of the Iterative Greedy algorithm to solve the Graph Coloring Problem. First, the basic principle of the NRPA algorithm is introduced. Then, NRPA is extended by using mixed sorting, destruction and reconstruction, and the Diversity-NRPA algorithm is proposed, which improves the diversity of the algorithm. Finally, Diversity-NRPA is applied to solve the Graph Coloring Problem by combining it with the knowledge of the graph theory field. We evaluate the performance of Diversity-NRPA on DIMACS, a well-known graph benchmark instance, and compare it with traditional graph coloring algorithms. The experimental results show that the Diversity-NRPA algorithm can achieve excellent performance in both solution quality and search efficiency in solving the Graph Coloring problem. [ABSTRACT FROM AUTHOR]

Nowadays, obtaining information from images is one of the main ways in which people obtain information. However, the images are affected by the acquisition of hardware equipment and transmission technology, which may result in the loss of certain data and the resolution to be reduced. How to recover low-resolution images into high-resolution images has grown to be a popular area of research for image processing. Deep learning techniques can discover more expressive features through adaptive learning from the dataset. However, there are problems with too many deep network parameters, blurred reconstructed images, and obvious human traces. To address these issues, this essay suggests the Deep Recursive Residual Subpixel Wasserstein Generative Adversarial Network (DRRSWGAN) using Deep Recursive Residual Subpixel Network (DRRSN) in the generative network to solve the problems of a large number of network parameters, relatively smooth reconstructed images and the presence of human traces. The Wasserstein Generative Adversarial Network (WGAN) is used in the discriminative network to improve the discriminative network of the GAN for the problem of training instability. Removing the Sigmod layer from the network, optimizing the loss function, and using the RMSProp algorithm for gradient optimization, allows the network to be more stable during training and better image reconstruction. The outcomes of the trial indicate this method improves both PSNR and SSIM metrics, and the image reconstruction results are better in terms of subjective perception. [ABSTRACT FROM AUTHOR]

[Display omitted] • Review of the knowledge structure using CiteSpace, VOSviewer, and SciMAT. • Visualizing the evolution of traffic collision reconstruction: the field is developing. • Analyzing co-word, co-citation, and major themes in traffic collision reconstruction. • Authors and institutions need more collaboration. • Head injury and uncertainty are hotspots: interdisciplinary research is the future. Traffic collisions are incidents with high fatality rate which generate billions of US dollars of loss worldwide each year. Post-collision scene reconstruction, which involves knowledge of multiple disciplines, is an important approach to restore the traffic collision and infer the cause of it. This paper uses software CiteSpace, VOSviewer, and SciMAT to conduct a visualization study of knowledge mapping on the literature of traffic collision scene reconstruction from 2001 to 2021 based on the Web of Science database. Knowledge mapping is a cutting-edge research method in scientometric, which has been widely applied in medicine and informatics. Compared with traditional literature review, knowledge mapping with visual techniques identifies hot keywords and key literature in the field more scientifically, and displays them in schematic diagrams intuitively which allows to further predict potential hotspots. A total of 803 original papers are retrieved to analyze and discuss the evolution of the field in the past 20 years, from macro to micro, in term of background information, popular themes, and knowledge structure. Results indicate the number of publications in this field is limited, and collaborations among authors and among institutions are insufficient. In the meantime, mappings imply the top three hot themes being scene reconstruction, computer technology, and injuries. The introduction of AI related technologies, such as neural networks and genetic algorithms, into collision reconstruction would be a potential research direction. [ABSTRACT FROM AUTHOR]

In the process of urban development, a large number of industrial heritages are in needs of being renovated as a result of the failure of meeting new needs. However, there has not yet formed an innovative and sustainable update design process. Based on the analysis of the problems existing in the renewal and reconstruction of industrial heritages, this paper, hereinafter, will establish an innovative design system by way of introducing extenics and BIM and combining the formalized and innovative advantages of extenics and the characteristics of BIM: information storage, processing and visual programming. In combination with the design of practical cases, this essay conducted the system, to verify its rationality and effectiveness; to discover limitations in its operation, and as the ultimate purpose, to propose directions for further research. [ABSTRACT FROM AUTHOR]

• Interpretable network design: NEFGN incorporates physical models to make the denoising network more interpretable. By grounding the network design in physical principles, it becomes easier to understand the underlying processes of noise reduction. • The NEFGN method proposed in this paper combines the advantages of traditional denoising and deep learning denoising to provide better interpretability and stronger noise reduction ability. • Construction of noise optimization model stage: By learning the parameters of the adaptive weighted total variation regularization model, NEFGN can approximate the noise level in NDCT images, so as to remove the noise in LDCT images more accurately. In addition, the study introduced noise level estimation maps as a means to accurately control the balance between noise reduction and detail retention for optimal visualization and image quality. • Construction of Gaussian curvature edge detection model stage: NEFGN uses Gaussian curvature guidance to construct an edge detection network to directly predict clear edges from LDCT images. This helps to preserve important edge details and avoid image blurring or information loss. And an edge detector for making edge labels is proposed, which can better deal with the noise reduction task of low-dose CT images. • Comprehensive experimental evaluation: The proposed method was evaluated using the "NIH-AAPM-Mayo Clinic Low Dose CT Grand Challenge" dataset, piglet dataset, and clinical dataset. The experimental results demonstrate that NEFGN effectively enhances the diagnostic performance of CT images by restoring the LDCT image structure with constrained image details. • Superior performance: NEFGN outperforms other denoising algorithms in terms of both quantitative measures and visual quality. This makes it a promising candidate for clinical applications, where accurate interpretation of CT images is crucial for diagnosis and treatment planning. The denoising task of low-dose CT images is a highly complex and uncertain inverse problem. Previous studies have primarily relied on convolutional neural network to reduce noise by learning the mapping from LDCT images to normal dose CT images. However, simply increasing the network depth alone is not an optimal choice due to the limited performance improvement and significant computational cost. In contrast, integrating prior knowledge of images with a model to assist in image reconstruction is a more efficient approach. This study proposes a new framework for denoising LDCT images, named Noise-Optimized Edge Feature Guided Network (NEFGN). The task of NEFGN is to integrate the noise optimization model of adaptive weighted total variation expansion, the edge detection model guided by Gaussian curvature, and image reconstruction into an end-to-end CNN framework. In order to achieve this goal, the noise optimization model is first constructed by learning the parameters in the adaptive weighted total variation regularization model to approximate the noise level in the NDCT image. The edge detection network is constructed using Gaussian curvature, predicting clear edges directly from the noise image. Finally, under the guidance of the noise optimization model and the edge detail model, NEFGN is more capable of suppressing artifact noise, demonstrating good accuracy and robustness, and can restore finer details. Numerous experimental studies demonstrate that the NEFGN denoising framework effectively restores the structure of LDCT images with limited image details and outperforms other methods in terms of performance. [ABSTRACT FROM AUTHOR]

[Display omitted] • Develop a style-embedding representation learning (SeRL) for unsupervised and unpaired abdomen CT and MR image translation. • The synthetic CT images from MR modality can be processed by the CT-pretrained models. • A patch-based corrosion data augmentation is applied to increase the amount and diversity of our dataset. • Early stage Hepatocellular Carcinoma patients with MR scans can be potentially screened by both our SeRL and CT-pretrained models. Hepatocellular carcinoma (HCC) is the third leading cause of cancer-related deaths worldwide, and the fastest-growing cause of cancer deaths in the United States. Computed tomography (CT) and magnetic resonance (MR) imaging are the key image-based examination methods for HCC patients. Patients may alternatively undergo CT and MR exams given a radiation dose of the former and the cost of the latter. Due to organ representation differences between CT and MR, it's a great challenge to detect the tumors from a time series of data consisting of both CT and MR by a single model. The annotations for these time series data will consume much time and labor for the physicians. Thus, we propose our style-embedding representation learning (SeRL) for unsupervised and unpaired abdomen CT and MR translation. Different from current medical image translation models, the style-representation information from real CT and real MR images has been embedded in the translation process to bypass some local minima during the convergence process and improve the synthesis results. Our patch-corrosion augmentation method enhances the style-embedding representation learning by bringing more diversity to the training data. Combined with the self-attention module, SeRL eliminates the noise caused by low grayscale pixel values during translation. Results on the unpaired HCC patient's CT and MR images show that our proposed SeRL is able to generate high quality CT images from MR ones. Evaluations such as Frechet Inception Distance (FID), Sliced Wasserstein Distance (SWD), and liver segmentation dice score are utilized to demonstrate our advantages over other state-of-the-art unsupervised methods. [ABSTRACT FROM AUTHOR]

Magnetic Resonance Imaging (MRI) is a crucial tool for quantitative image analysis and clinical diagnosis, providing detailed anatomical images to assist in the detection of various abnormalities. However, the widespread use of MRI is hindered by the challenges associated with long sampling periods and low-resolution image processing, prompting the development of various traditional methods to address these limitations. In recent times, advanced Deep Learning (DL) techniques have been applied to tackle the inverse problem of reconstructing MRI images from undersampled data. These DL models have demonstrated substantial improvements in terms of image reconstruction performance, cost-effectiveness, and reduced acquisition time, offering significant potential for further enhancements in this domain. This study introduces a novel DLGAN (Deep Learning Generative Adversarial Network) model comprising two sub-GAN modules tailored for specific datasets. Each module incorporates DLGAN (Generator Block, GB) and DLGAN (Discriminator Block, DB) blocks, strategically designed to regenerate MR images from K Space data. These blocks effectively leverage information from both the ground truth and KSpace characteristics, resulting in enhanced image reconstruction performance while also reducing complexity and improving overall efficiency. The DLGAN model effectively addresses common issues such as artefact removal and vanishing-gradient problems through the extraction of hierarchical features. To evaluate the model's effectiveness, comprehensive experiments were conducted, utilizing metrics such as Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM). The experimental findings demonstrate that the proposed DLGAN model outperforms the most recent designs in MRI image reconstruction, showcasing its potential for significant advancements in clinical diagnosis and quantitative image analysis. • DL-GAN-based MRI reconstruction algorithm for medical images is developed. • Dual reconstruction blocks are used to reconstruct images with reduced complexity. • Proposed methods PSNR and SSIM are higher for reconstructing medical images. [ABSTRACT FROM AUTHOR]

• Magnetic resonance images are not ordinary natural images. • The K-space is rich in information features between nodes. • Graph neural networks can extract the information features between nodes. • Combination of dual-domain features can improve the quality of reconstructed images. To recover the missing high-frequency components in low-resolution images by activating nodes in the high-frequency region of k-space. In addition, combining the information features in the k-space domain with the structural features in the image domain to generate high-quality super-resolution images. We propose a novel Dual-Domain cascaded Super-Resolution Network (DDSRNet), which shifts the starting point of the network to k-space. DDSRNet combines the ideas of GraphSAGE, Swin-Unet, and Hybrid Attention Transformer, to achieve complementary advantages of the k-space domain and the image domain. The k-space domain network activates zero-filled nodes in high-frequency regions and generates multi-order information features. The image domain network performs shallow, deep and gradient feature extraction to obtain the high-level representation of dual-domain hybrid features. In addition, to better utilize the features of both domains, we construct a domain interaction pool to facilitate cross-domain feature transfer and improve the efficiency of feature fusion. Exhaustive experiments are conducted on both public dataset and real scanning dataset. Compared with state-of-the-art algorithms, DDSRNet has the best numerical evaluation results (the average PSNR and SSIM improvements are more than 0.1 dB and 0.0078) and visual perception. In addition, DDSRNet enables high-quality 2x SR MRI with a reduced number of excitations, leading to increased low-field imaging speed. DDSRNet shows excellent performance in high- and low-field SR tasks and has the potential to be a powerful tool for clinical applications. The proposed method is of great practical significance for achieving fast and high-quality magnetic resonance imaging. [ABSTRACT FROM AUTHOR]

In the medical field, Medical diagnosis using Computed tomography (CT) has become increasingly popular due to their non-invasive approach and quick overall turnaround time. 3D visualization for CT Colonography involves the assessment and diagnosis of a patient to find the presence of cancerous polyps in the colon, by taking a Computed Tomography scan of the patient, and evaluating the reports. This technique reduces evaluation time by allowing the doctors themselves to analyze the CT scans without the need for a radiologist to generate an initial report. This technique also avoids an invasive procedure on the patient. This paper gives the insightsof developing computer aided system for the detection of Polyps in CT Colonography images using the principles of Image Processing and the Deep Learning, specifically an ensemble of Convolutional Neural Networks using GoogleNet architecture and 3D reconstruction of the same. The accuracy achieved by the proposed system for region classification, region 1 polyp detection and region 2 polyp detection are 98.75%, 93.75% and 94.03% respectively, and their F1 Scores are 0.88, 0.82 and 0.84 respectively. [ABSTRACT FROM AUTHOR]

Diffuse optical tomography (DOT) is promising to investigate functional and structural information in biological tissue by involving the forward problem and inverse solution. The work within this study develops an image reconstruction algorithm based on frequency-domain DOT using multi-frequency. In the forward problem, numerical simulation performs a diffusion equation to compute the light radiance distribution inside the model geometry with the assumed light intensity and optical properties. While in inverse solution, the absorption and reduced scattering coefficients are reconstructed utilizing light boundary data. To test the algorithm, several simulation cases, such as different contrast ratios between background and inclusion, inclusion size, as well as single and multi-frequency using one to three frequencies each and simultaneously are implemented. For verifying the image reconstruction results of multi-frequency, an objective numerical assessment, namely contrast and size detail is applied with a threshold value to specify the visible inclusion inside the geometry. Reconstructed images indicate that the effectivity of two frequencies in multi-frequency with 100 and 150 MHz has similar results as in single-frequency with 200 MHz. Therefore, the more number of frequencies may not enhance the contrast and image quality. [ABSTRACT FROM AUTHOR]

Current reconstruction of 3D images from DICOM (Digital Imaging and Communications in Medicine) files requires strict supervision for the reconstructed images to have the same metadata including slice thickness, spacing between slices, and image resolution. We propose an algorithm for reconstructing 3D images based on medical images in the DICOM format with varied metadata and resizing the 3D images while preserving the annotations. The 3D image resizing may facilitate processing because most current systems cannot handle the huge 3D image data sizes. After resizing the 3D images, the original annotations that can be used as ground truths to train and evaluate machine learning method are preserved by projection. Experimental results show that the proposed method can handle various DICOM files and correctly project annotations onto the resized image. [ABSTRACT FROM AUTHOR]

This paper presents an adaptive and robust method for image copyright protection to control ownership and prevent unauthorized usage of image property. The image is transformed into a wavelet domain using a multi-level of lifting wavelet transform; the lowest frequency band is globally divided into several concentric rectangles to withstand cropping. Pixels of a user-selected rectangle are designated for embedding flag bits to protect against rotational attacks. The proposed embedding operation takes into consideration the coefficient of high energy to be identified as potential positions for the embedding process. This procedure reduces the total error presented by the embedding process and protects hidden data against noise and jpeg-compression attacks. A reversible scrambling is applied to a pre-defined region from the watermarked image to prevent unauthorized users from attaining a high quality watermarked image. Experimental results show that the proposed algorithm achieves a range of PSNR (44–50) dB and a normalized correlation range of (1–0.9) for the watermarked image and reconstructed logo. Results also demonstrate the robustness against various attacks, such as rotation, cropping, noise adding, and JPEG-compression. [ABSTRACT FROM AUTHOR]

Computed tomography (CT) is a non-destructive evaluation technique to know the internal structure of the objects under scan. It has numerous applications in engineering as well as in the medical field. The prime objective of this manuscript is to reduce the radiation dose to the object under scanning and to reconstruct the low-noise CT images, even in limited-view projections data scenarios. The present manuscript proposes a novel priority-based self-guided serial-parallel hybrid genetic algorithm for low-dose CT reconstruction. The current algorithm combines serial and parallel processing elements to reduce the loss of diversity in the population and increase the convergence rate. The proposed algorithm uses a novel priority-based self-guided competition operator. It consists of two approaches for maintaining exploration and exploitation tradeoffs. Here, the first approach uses serial processing, whereas the second approach uses parallel processing. The algorithm also uses priority-based hybrid crossover (PHCO) and adaptive priority-based mutation operator (PMO) to generate the most suitable offspring in every generation. Experimental results reveal that the presented algorithm produces satisfactory results with limited-view projections. The presented algorithm also outperforms other limited-view CT reconstruction algorithms. • The proposed algorithm reconstructs low-noise CT images with limited view projections. • The algorithm reduces the radiation dose and cost per scan. • The presented algorithm proposes a self-guided competition operator. • The self-guided competetion operator ensure the diversity of the solution space. • The presented algorithm supports a valuable diversity of the solution space. • The algorithm preserves the texture and structure properties of regenerated images. • The algorithm provides a high convergence rate to find an optimal solution. • The proposed algorithm is appropriate for low-dose CT reconstruction. [ABSTRACT FROM AUTHOR]

Camera traps are widely used for wildlife monitoring and making informed conservation and land-management decisions, but the resulting 'big data' are laborious to process. Deep learning-based methods have been adopted for wildlife detection in camera traps. However, these methods detect large mammals in uncomplicated scenes, where powerful deep-learning models work effectively. Few studies have been conducted to develop artificial intelligence for recognizing wild birds that live in complicated field scenes with protective colors and small sizes. Here we used a dataset of 9717 images from 15 bird species based on camera traps to test 8 object detection algorithms (Faster RCNN, Cascade RCNN, RetinaNet, FCOS, RepPoints, ATSS, Deformable-DETR, and Sparse RCNN) and assess their performance. We also explored the effect of different backbones on model accuracy. Among them, the Cascade RCNN model performs best, with a mAP of 0.693 in model capabilities. Models perform differently in certain species, and backbones significantly affect the accuracy of the model. Cascade RCNN utilizing the Swin-T backbone is the best-performing combination, with a mAP of 0.704. This study could help researchers identify birds efficiently and inspires research on wildlife recognition in complex ecological settings. • Constructed a dataset of 15 bird species using Camera Traps. • Employed deep learning to identify birds in complex backgrounds. • Cascade RCNN models outperformed other object detection models. • Backbones choice significantly impacted bird recognition accuracy. • Deep learning shows potential for wildlife recognition in complex scenarios. [ABSTRACT FROM AUTHOR]

Within powder bed fusion of metals by a laser-beam (PBF-LB/M) spatter particles evolve due to the entrainment of powder particles and their ejection along the vapor plume. Motion properties of the evolving spatter particles are directly linked to the process zone topography, qualifying them as an interesting process signature for in-situ sensing. This work shows an industrially applicable sensing approach which allows the determination of three-dimensional spatter positions in real-time. A camera setup is derived reducing ambiguity within the reconstruction at high framerates. The presented parallel image processing and reconstruction algorithm allows for the determination of three-dimensional spatter coordinates in real-time. [ABSTRACT FROM AUTHOR]

In this paper, we propose a non-convex regularization magnetic resonance imaging (MRI) reconstruction model via the smoothly clipped absolute deviation (SCAD) penalty function, which can effectively improve the fitting performance and prevent systematic underestimation compared with the classical total variation (TV) regularization. Then, we choose the alternating direction method of multipliers (ADMM) algorithm to solve the non-convex regularization model. The experiment results show that the efficiency of the proposed model and algorithm in comparison with some other typical methods. [ABSTRACT FROM AUTHOR]

The sparsity adaptive matching pursuit (SAMP) algorithm reconstructs the signal through multiple iterations with a fixed stepsize, which consumes too much time. To solve this problem, a variable stepsize sparsity adaptive matching pursuit (VSSAMP) algorithm is proposed. The sparsity is pre-estimated, and then the variable stepsize SAMP is used to reconstruct the signal. Different from the SAMP algorithm, VSSAMP algorithm determines the stepsize according to the energy difference of two consecutive residues. The experiment results show that the reconstruction probability of VSSAMP algorithm is slightly slower than SAMP algorithm in one-dimensional signal, and the reconstruction quality of VSSAMP is basically equal to SAMP algorithm in tow-dimensional image. Most notably, the reconstruction time of VSSAMP algorithm is significantly lower than SAMP in reconstruction of one-dimensional signal and two-dimensional image. [ABSTRACT FROM AUTHOR]

Sparse representation models have been widely used in single-image super-resolution reconstruction. The construction of dictionaries is especially important in these models. Basically, approaches of dictionary construction in sparse representation can be divided into two categories: analytical and learning-based approaches. Analytical approaches are effective and fast, but they are unable to fit different types of data; learning-based approaches are adaptive, but their implementation takes a significant amount of time. In this study, an image super-resolution reconstruction algorithm based on double-sparsity dictionaries is proposed. The algorithm combines the efficiency of analytical approaches and adaptability of learning-based approaches. In addition to the sparsity prior, the non-local self-similarity prior is also considered in the algorithm. Non-local means filtering is used to be the constraints on regularized super-resolution reconstruction procedures, which improves the quality of super-resolution reconstruction results further, while the runtime of the algorithm is still acceptable. Experimental results demonstrate the advantage of the proposed algorithm. [ABSTRACT FROM AUTHOR]

Copyright of Journal of Harbin Institute of Technology. Social Sciences Edition / Haerbin Gongye Daxue Xuebao. Shehui Kexue Ban is the property of Harbin Institute of Technology and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

In compressed sensing (CS), the traditional matching pursuit algorithms have a narrow adaptability to the sparsity and a higher time complexity. To expand the adaptability to sparsity and reduce the time complexity, a regularized-subspace pursuit (R-SP) algorithm is proposed. The regularization rule of the regularized orthogonal matching pursuit (ROMP) algorithm and the backtracing mechanism of the subspace pursuit (SP) algorithm are used to improve the accuracy of atom selection. The results of experiment show that in one-dimensional signal, the reconstruction probability of each algorithm is almost the same when the sparsity K is small. However, when the sparsity K increases, the R-SP algorithm has higher reconstruction probability and obvious advantages of reconstruction time. In two-dimensional images, the reconstruction performance of R-SP algorithm is slightly worse than ROMP algorithm. What's more, the R-SP algorithm widens the range of sparsity K, shortens the reconstruction time and achieves the complementary advantages when compared with other algorithms. [ABSTRACT FROM AUTHOR]

Low-light image enhancement is a key technique to overcome the quality degradation of photos taken under scotopic vision illumination conditions. The degradation includes low brightness, low contrast, and outstanding noise, which would seriously affect the vision of the human eye recognition ability and subsequent image processing. In this paper, we propose an approach based on deep learning and Retinex theory to enhance the low-light image, which includes image decomposition, illumination prediction, image reconstruction, and image optimization. The first three parts can reconstruct the enhanced image that suffers from low- resolution. To reduce the noise of the enhanced image and improve the image quality, a super- resolution algorithm based on the Laplacian pyramid network is introduced to optimize the image. The Laplacian pyramid network can improve the resolution of the enhanced image through multiple feature extraction and deconvolution operations. Furthermore, a combination loss function is explored in the network training stage to improve the efficiency of the algorithm. Extensive experiments and comprehensive evaluations demonstrate the strength of the proposed method, the result is closer to the real-world scene in lightness, color, and details. Besides, experiments also demonstrate that the proposed method with the single low-light image can achieve the same effect as multi-exposure image fusion algorithm and no ghost is introduced. [ABSTRACT FROM AUTHOR]

In this study, we present a novel deep learning network model, EAswin-unet, for accurately segmenting COVID-19 lesions in chest CT images. Our model adopts a fusion strategy that combines edge-weighted attention features and Swin-unet pure attention model features to increase the weight of edge recognition and improve segmentation accuracy. Compared to the inf-net network, our model achieved a significant increase of 0.055 in Dice score and 0.024 in sensitivity, while reducing the number of model parameters by 11770. To further improve the accuracy of our model, we also employed a hybrid semi-supervised algorithm strategy to use labeled data to correct model training and extract non-edge local information. Our proposed model was evaluated on a dataset of 100 labeled chest CT images from COVID-19 patients and 1600 unlabeled data, achieving a Dice similarity coefficient (DSC) of 0.737, sensitivity of 0.716, and specificity of 0.929, surpassing state-of-the-art methods. These results demonstrate the effectiveness of our proposed model in accurately segmenting COVID-19 lesions in chest CT images, providing an important tool for radiologists and clinicians in the diagnosis and treatment of COVID-19 patients. • We propose the EAswin-Unet model based on edge attention, which enables us to obtain better position information and edge details of lesions. • By combining attention and pixel-weighted loss function, we improve the model's focus on local information during the training process and increase the accuracy of COVID-19 lesion segmentation. • We propose a replacement-based semi-supervised strategy for training the model. This strategy helps alleviate the shortcomings of the original model in small data labels, such as inadequate data, low model fitting ability, and poor generalization ability during training. [ABSTRACT FROM AUTHOR]

• Propose an edge-preserving total nuclear variation for sparse reconstruction. • Propose a method to dynamically adjust the intensity of gradient diffusion. • Extensive experiments have demonstrated the effectiveness of the proposed method. Currently, slow scanning speed has become a technical bottleneck limiting the development of electron paramagnetic resonance imaging (EPRI), a novel oxygen imaging modality. Image reconstruction from sparse-view projections is an important imaging configuration to achieve fast scanning in EPRI. However, EPR images reconstructed from sparse-view projections using traditional analytic algorithms suffer from sparse artifacts, causing severe image degradation. In this work, we aim to use an optimization-based approach to achieve high-precision sparse reconstruction for EPRI. Methods. Inspired by the success of total nuclear variation (TV N) for denoising multi-channel spectral CT images and adaptive-weighted total variation (awTV) for improving the edge-preserving performance of TV, we proposed an edge-preserving total nuclear variation (EPTV N) minimization algorithm for sparse reconstruction in EPRI. The EPTV N combines the advantage of TV N in modelling structural similarity and low-rank prior between multi-channel images for denoising with the advantage of awTV for edge-preserving, which can better protect the edge structure of images while suppressing sparse artifacts. Results. We designed two numerical phantoms with different properties, a piecewise-constant phantom and a gradient phantom, and a physical phantom, and then performed inverse crime, simulation study and real-data study. Experimental results show that the EPTV N minimization algorithm can reconstruct more accurate EPR images from sparse-view projections for both the numerical and physical phantoms. Significance. The proposed method can effectively suppress sparse artifacts and protect edge structures of reconstructed images, thus achieving high-precision sparse-view reconstruction for EPRI. The insights of this work can also be applied to multi-channel image denoising tasks. [ABSTRACT FROM AUTHOR]

Although a number of license plate recognition (LPR) systems have become significantly advanced, they are still far from producing ideal images. They are still fiddled with low-resolution (LR) and motion blur, which are two of the most common problems in images extracted from automobile driving environment. In order to address this issue, we present a novel LPR method that processes LR and motion blurred images from dash cams. We propose a unique framework, AFA-Net (Adaptive Feature Attention Network) organized by image pre-restoration, feature composition and image reconstruction modules. The proposed AFA-Net can generate a clear restoration image for robust LPR performance with the images obtained from dash cams in an unconstrained environment. Furthermore, we explore the novel problem, Joint-IRLPRNet (Joint-Image Restoration and License Plate Recognition Network), that simultaneously address image restoration (i.e. SR and deblurring) and LPR in an end-to end trainable manner. Moreover, we introduce a dataset called LBLP (LR and blurred license plate (LP)). The dataset is composed of 2,779 LR and motion blurred cropped LP images, extracted from unconstrained dash cams. The experimental results on LBLP dataset indicate that AFA-Net achieves 15.28% improvement in recognition accuracy, 6.47% in sequence similarity, and 3.89% in character similarity, compared to the traditional LPR model with image restoration model. Moreover, Joint-IRLPRNet can be more effective results than AFA-Net. • Our AFA-Net can generate a clear restoration image for robust LPR performance. • We construct a dataset called LBLP extracted from unconstrained dash cams. • AFA-Net can be more effective results than traditional LPR models. • We present a multi-tasking paradigm by handling image restoration issues with LPR. [ABSTRACT FROM AUTHOR]

The challenges posed by high pixel resolution and complex backgrounds in UAV remote sensing images have hindered the effective feature extraction and precise bounding box regression for small targets. In response to these challenges, numerous detection methods based on deep learning have emerged in recent years. Despite their advancements, these methods have not fully addressed the demand for accurate identification of small targets in remote sensing images. This paper introduces DSAA-YOLO, a novel algorithm designed for small target detection in UAV remote sensing images. Firstly, a new data augmentation strategy, termed Super Resolution Data Augment (SRDA), is proposed, which integrates the concept of image super-resolution to enrich the dataset while preserving data quality. Furthermore, a Dense Residual-based Super-Resolution module (DRSR) is introduced to enhance the resolution of small targets that have undergone quality degradation due to transformations. Subsequently, an Information Alignment Feature Enhancement Module (IAFE) is proposed to maximize the extraction of original features from the image. Finally, based on the improved Multi-Objective Grey Wolf Optimization (MOGWO), a novel dynamic anchor regression strategy termed Multi-Object Golf Dynamic Anchor (MGDA) is devised to generate more precise bounding boxes. The proposed DSAA-YOLO algorithm demonstrates significant improvements over current state-of-the-art methods in terms of widely recognized metrics including mAP, and AP50 on the VisDrone dataset. [ABSTRACT FROM AUTHOR]

Compressed sensing in image reconstruction has attracted attention and many studies are proposed. As we know, adding prior knowledge about the distribution of the support on the original signal to CS can improve the quality of reconstruction. However, it is still difficult for a recovery framework adjusts its strategy for exploiting the prior knowledge efficiently according to the current estimated signals in serial iterations. With the theory of information geometry, we propose an adaptive strategy based on the current estimated signal in each iteration of the recovery. We also improve the performance of existing algorithms through the adaptive strategy for exploiting the prior knowledge according to the current estimated signal. Simulations are presented to validate the results. In the end, we also show the application of the model in the image. [ABSTRACT FROM AUTHOR]

In the 5G era, the communication networks tend to be ultra-densified, which will improve the accuracy of indoor positioning and further improve the quality of positioning service. In this study, we propose an indoor three-dimensional (3D) dynamic reconstruction fingerprint matching algorithm (DSR-FP) in a 5G ultra-dense network. The first step of the algorithm is to construct a local fingerprint matrix having low-rank characteristics using partial fingerprint data, and then reconstruct the local matrix as a complete fingerprint library using the FPCA reconstruction algorithm. In the second step of the algorithm, a dynamic base station matching strategy is used to screen out the best quality service base stations and multiple sub-optimal service base stations. Then, the fingerprints of the other base station numbers are eliminated from the fingerprint database to simplify the fingerprint database. Finally, the 3D estimated coordinates of the point to be located are obtained through the K-nearest neighbor matching algorithm. The analysis of the simulation results demonstrates that the average relative error between the reconstructed fingerprint database by the DSR-FP algorithm and the original fingerprint database is 1.21%, indicating that the accuracy of the reconstruction fingerprint database is high, and the influence of the location error can be ignored. The positioning error of the DSR-FP algorithm is less than 0.31 m. Furthermore, at the same signal-to-noise ratio, the positioning error of the DSR-FP algorithm is lesser than that of the traditional fingerprint matching algorithm, while its positioning accuracy is higher. [ABSTRACT FROM AUTHOR]

The ancient pottery is often found in archaeological sites in a broken state, especially when those pieces of unknown organisms and irregular fragments, may take years of hard work, especially in the case of loss of some pieces or require hard work and experienced archaeologists. So this problem is divided into two major tasks the first of which is the Classification of Archaeological Fragments into similar groups (CAF) and the second one is the Reconstruction of each group into the original Archaeological Objects (RAO). To solve this problem, a method has been proposed, which exploits the color and texture properties of the surfaces of the fragments. Furthermore, the reconstruction of archaeological fragments in 3D geometry is an important problem in pattern recognition. Therefore, this research has implemented the algorithms to reconstruct real datasets using Neural Networks. The challenge of this work is to reconstruct the objects without previous knowledge about the part that should start the assembly; this greatly helps to avoid the presence of gaps created due to missing artifact fragments. The study utilizes the geometric features of the fragments as important features to reconstruct the objects by classifying their fragments using a Neural Network model. [ABSTRACT FROM AUTHOR]

Low-rank representation methods already achieve many applications in the image reconstruction. However, for high-gradient image patches with rich texture details and strong edge information, it is difficult to find sufficient similar patches. Existing low-rank representation methods usually destroy image critical details and fail to preserve edge structure. In order to promote the performance, a new representation-based image super-resolution reconstruction method is proposed, which combines gradient domain guided image filter with the structure-constrained low-rank representation so as to enhance image details as well as reveal the intrinsic structure of an input image. Firstly, we extract the gradient domain guided filter of each atom in high resolution dictionary in order to acquire high-frequency prior information. Secondly, this prior information is taken as a structure constraint and introduced into the low-rank representation framework to develop a new model so as to maintain the edges of reconstructed image. Thirdly, the approximate optimal solution of the model is solved through alternating direction method of multipliers. After that, experiments are performed and results show that the proposed algorithm has higher performances than conventional state-of-the-art algorithms in both quantitative and qualitative aspects. [ABSTRACT FROM AUTHOR]

Low-rank representation methods already achieve many applications in the image reconstruction. However, for high-gradient image patches with rich texture details and strong edge information, it is difficult to find sufficient similar patches. Existing low-rank representation methods usually destroy image critical details and fail to preserve edge structure. In order to promote the performance, a new representation-based image super-resolution reconstruction method is proposed, which combines gradient domain guided image filter with the structure-constrained low-rank representation so as to enhance image details as well as reveal the intrinsic structure of an input image. Firstly, we extract the gradient domain guided filter of each atom in high resolution dictionary in order to acquire high-frequency prior information. Secondly, this prior information is taken as a structure constraint and introduced into the low-rank representation framework to develop a new model so as to maintain the edges of reconstructed image. Thirdly, the approximate optimal solution of the model is solved through alternating direction method of multipliers. After that, experiments are performed and results show that the proposed algorithm has higher performances than conventional state-of-the-art algorithms in both quantitative and qualitative aspects. [ABSTRACT FROM AUTHOR]

The growing quality of computer graphics makes it difficult to classify real photos from rendered images. At the same time, a high-quality classifier that separates real-world photos from the images created by computer graphics allow solving many applied problems: from automatic evaluation of the quality of the image to detecting manually modified photos. The main directions of research in this area are associated with the identification of distinctive features in the images for use in classification models. In this paper, we provide experimental studies of various classification models and groups of features on the image dataset from Vkontakte social network. We show that the best choice from a variety of features for solving the classification problem is the combination of Haralick features, Bag of Visual Words (BoVW) with Speeded-Up Robust Features (SURF) descriptors and Local Binary Pattern (LBP) histograms. In addition, we show that Gradient Boosting model being trained on the specified combination of feature groups outperform other machine learning models with 90% accuracy. [ABSTRACT FROM AUTHOR]

Data Security is the most challenging area in Internet communication, where most of the secret sharing schemes are proposed for binary images. But still it lacks in providing security for data communication, especially in image transmission. Traditional visual cryptography scheme generate meaningless diwies and the reconstruction phase leads to quality degradation over the secret image. In this work, an intelligent two dimensional secret share construction scheme is proposed. A secret image is expanded into n diwies with the choice of scheme selection. By Stacking all the qualified diwies to revert the secret image without content loss and less than s* - 1 shares could not reveal any information about the secret image. The experimental results emphasize that the proposed secret share scheme is highly secured for image transmission. [ABSTRACT FROM AUTHOR]

Considering that high-dose X-ray radiation during CT scans may bring potential risks to patients, in the medical imaging industry there has been increasing emphasis on low-dose CT. Due to complex statistical characteristics of noise found in low-dose CT images, many traditional methods are difficult to preserve structural details effectively while suppressing noise and artifacts. Inspired by the deep learning techniques, we propose a densely connected residual network (DCRN) for low-dose CT image noise cancelation, which combines the ideas of dense connection with residual learning. On one hand, dense connection maximizes information flow between layers in the network, which is beneficial to maintain structural details when denoising images. On the other hand, residual learning paired with batch normalization would allow for decreased training speed and better noise reduction performance in images. The experiments are performed on the 100 CT images selected from a public medical dataset—TCIA(The Cancer Imaging Archive). Compared with the other three competitive denoising algorithms, both subjective visual effect and objective evaluation indexes which include PSNR, RMSE, MAE and SSIM show that the proposed network can improve LDCT images quality more effectively while maintaining a low computational cost. In the objective evaluation indexes, the highest PSNR 33.67, RMSE 5.659, MAE 1.965 and SSIM 0.9434 are achieved by the proposed method. Especially for RMSE, compare with the best performing algorithm in the comparison algorithms, the proposed network increases it by 7 percentage points. [ABSTRACT FROM AUTHOR]

In the production of industrial fabric, it needs automatic real-time system to detect defects on the fabric for assuring the defect-free products flow to the market. At present, many visual-based methods are designed for detecting the fabric defects, but they usually lead to high false alarm. Base on this reason, we propose a Sobel operator combined with patch statistics (SOPS) algorithm for defects detection. First, we describe the defect detection model. mean filter is applied to preprocess the acquired image. Then, Sobel operator (SO) is applied to deal with the defect image, and we can get a coarse binary image. Finally, the binary image can be divided into many patches. For a given patch, a threshold is used to decide whether the patch is defect-free or not. Finally, a new image will be reconstructed, and we did a loop for the reconstructed image to suppress defects noise. Experiments show that the proposed SOPS algorithm is effective. [ABSTRACT FROM AUTHOR]

Due to inherent physical and hardware limitations, 3D MR images are often acquired in the form of orthogonal thick slices, resulting in highly anisotropic voxels. This causes the partial volume effect, which introduces blurring of image details, appearance of staircase artifacts and significantly decreases the diagnostic value of images. To restore high resolution isotropic volumes, we propose to use a convolutional neural network (CNN) driven by patches taken from three orthogonal thick-slice images. To assess the validity and efficiency of this postprocessing approach, we used 1x1x1 mm3-voxel brain images of different modalities, available via the well known BrainWeb database. They served as a high resolution reference and were numerically preprocessed to create input images of different slice thickness and anatomical orientation, for CNN training, validation and testing. The visual quality of reconstructed images was indeed superior, compared to images obtained by fusion of interpolated thick-slice images, or to images reconstructed with the CNN using a single input MR scan. The significant increase of objectively computed figures of merit, e.g. the Structural Similarity Image Metric, was also noticed. Keeping in mind that any single value of such quality metrics represents a number of psychophysical effects, we applied the CNN trained on brain images for superresolution reconstruction of synthetic and acquired blood vessel tree images. We then used the restored superresolution volumes for estimation of vessel radii. It was demonstrated that vessel radius values derived from superresolution images of simulated vessel trees are significantly more accurate than those obtained from a standard fusion of interpolated thick-slice orthogonal scans. Superiority of the CNN-based superresolution images was also observed for scanner-acquired MR scans according to the evaluated parameters. These three experiments show the efficiency of CNN-based image reconstruction for qualitative and quantitative improvement of its diagnostic quality, as well as illustrates the practical usefulness of transfer learning - networks trained on example images of one kind can be used to restore superresolution images of physically different objects. [ABSTRACT FROM AUTHOR]

Image acquisition in heterogeneous environments tends to lead to inadequate information and poor image quality. In order to improve the image quality of dynamic data in heterogeneous environments, an enhanced technology of dynamic data information based on Radon scale transformation is proposed. The gray histogram feature information parameters of images in heterogeneous environments are extracted. The feature quantities are fused and optimized in the central region of image clustering, and the multi-scale Retinex color feature components of dynamic data are extracted. Radon scale transformation is used to extract image centers in heterogeneous environments, enhance dynamic data of image centers, and improve image quality. The simulation results show that this method can enhance the dynamic data information of image centers in heterogeneous environments, and the output images have better imaging performance. The normalized correlation coefficient and peak signal-to-noise ratio (psnr) of the output images are higher than those of the traditional methods, which improves the peak signal-to-noise ratio (psnr) of the output image and improves the recognition performance of the image. [ABSTRACT FROM AUTHOR]

Copyright of Journal of Harbin Institute of Technology. Social Sciences Edition / Haerbin Gongye Daxue Xuebao. Shehui Kexue Ban is the property of Harbin Institute of Technology and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

An intelligent diagnosis method based on support vector machine (SVM) is proposed to quantitatively diagnose early weak faults of high-speed aero-engine's bearings. In order to achieve a better performance in contrast with conventional kernel function, a mixed kernel function is constructed and particle swarm optimization (PSO) is used to optimize kernel coefficients and other parameters. Experimental raw data is preprocessed by sparse decomposition and reconstruction method to remove noise in original signals, which can provide effective and reliable samples for SVM. In order to verify the validity of the proposed method, experiments on different fault types with different defect sizes of high-speed bearings working at 30000rpm are carried out. The results show that the accuracy of the proposed method is greatly improved compared with traditional SVM. The proposed method can not only distinguish different types of failure but also distinguish different degrees of the same fault pattern, which achieves a quantitative intelligent diagnosis of early weak faults in aviation's high-speed bearings. • A hybrid kernel function integrating other single kernel merits is proposed. • Particle swarm optimization (PSO) is proposed to select optimal kernel coefficients. • Stepwise orthogonal matching pursuit algorithm (StOMP) is utilized to singularize periodic impact signal. • The proposed method can effectively solve the problem of quantitative fault diagnosis of bearing with large DN value. [ABSTRACT FROM AUTHOR]

Meta-learning has recently been an emerging data-efficient learning technique for various medical imaging operations and has helped advance contemporary deep learning models. Furthermore, meta-learning enhances the knowledge generalization of the imaging tasks by learning both shared and discriminative weights for various configurations of imaging tasks during training. However, existing meta-learning models attempt to learn a single set of weight initializations of a neural network that might be fundamentally restrictive under the heterogeneous (multimodal) data scenario. This work aims to develop a multimodal meta-learning model for image reconstruction, which augments meta-learning with evolutionary capabilities to encompass diverse acquisition settings of heterogeneous data. Our proposed model called KM-MAML (K ernel M odulation-based M ultimodal M eta- L earning), has hypernetworks (auxiliary learners) that evolve to generate mode-specific (or context-specific) weights. These weights provide the mode-specific inductive bias for multiple modes by re-calibrating each kernel of the base network for image reconstruction via a low-rank kernel modulation operation. Furthermore, we incorporate gradient-based meta-learning (GBML) in the contextual space to update the weights of the hypernetworks based on different modes. The hypernetworks and the base reconstruction network in the GBML setting provide discriminative mode-specific features and low-level image features, respectively. We extensively evaluate our model for multi-contrast magnetic resonance image reconstruction considering the essential research directions in fastMRI for multimodal and rich transfer learning capabilities across various MRI contrasts. Our comparative studies show that the proposed model (i) exhibits superior reconstruction performance over joint training, other meta-learning methods, and various context-specific MRI reconstruction architectures, and (ii) better adaptation to 80% and 92% of unseen multi-contrast data contexts with improvement margins of 0.1 to 0.5 dB in PSNR and around 0.01 in SSIM, respectively. Besides, a representation analysis with U-Net as the base network shows that kernel modulation infuses 80% of mode-specific representation changes in the high-resolution layers. Our source code is available at https://github.com/sriprabhar/KM-MAML/. [Display omitted] • Multimodal meta-learning provides mode-specific inductive bias for multi-contrast MRI reconstruction. • Hypernetworks exhibit evolutionary capabilities to dynamically predict weights that modulate the reconstruction network weights. • Gradient-based meta-learning to optimize the kernel modulation hypernetworks in the contextual space. • Representational analysis to interpret the mode-specific weights of the hypernetworks. [ABSTRACT FROM AUTHOR]

In this paper, we propose Convolutional Neural Network models of Multivariate Regression (MRCNN) applied for the reconstruction of satellite images. Image reconstruction is an ill-posed inverse problem of computer vision. Generally, distortion in satellite images modeled as their convolution with an Airy pattern Point Spread Function (PSF) of circular symmetry and additive white Gaussian noise. The MRCNN training needs a few satellite images, to yield a reconstructed image during testing, with the goals of generalized and optimized restoration. Thus, it must manage a tradeoff between two objectives of optimization and generalization. The image to be restored is not provided to the network during training. The CNN has already proven its effectiveness in image classification and recognition but not thoroughly evaluated on multivariate regression and inverse problems of image processing and computer vision. Therefore, we experimented with different architectures of CNN feasible on a GPU embedded PC for an effective reconstruction of satellite images. The proposed network 5 comprised on three CNN layers and a dense layer with a novel data engineering procedure has shown better reconstruction of satellite images evaluated by peak-signal-to-noise-ratio. [ABSTRACT FROM AUTHOR]

In this paper, we introduced a novel image clustering method based on combination of the classical F uzzy C - M eans (FCM) algorithm and B acktracking S earch optimization A lgorithm (BSA). The image clustering was achieved by minimizing the objective function of FCM with BSA. In order to improve the local search ability of the new algorithm, an inertia weight parameter (w) was proposed for BSA. The improvement was accomplished by using w in the steps of the determination of the search-direction matrix of BSA and the new algorithm was named as w-BSAFCM. In order to show the effectiveness of the new algorithm, FCM was also combined with the general forms of BSA in the same manner and three benchmark images were clustered by utilizing these algorithms. The obtained results were analyzed according to the objective function and Davies-Bouldin index values to compare the performances of the algorithms. According to the results, it was shown that w-BSAFCM can be effectively be used for solving image clustering problem. [ABSTRACT FROM AUTHOR]

The ability to investigate 3D structure of biomolecules, such as proteins and viruses, is essential in biology and medicine. With the invention of super-bright X-ray free electron lasers (e.g. European XFEL and Linac Coherent Light Source (LCLS)) the Single Particle Imaging (SPI) approach allows to reconstruct 3D structures from many 2D diffraction images produced in the experiment by X-rays scattered on the biomolecule exposed in different orientations. In the same time, there are still many challenging problems in experiment setup, sample preparation and injection, which limit number and quality of obtained diffraction patterns and, consequently, limit achievable resolution of recovered 3D structure. However even with the current experimental limitations it is possible to reconstruct 3D structures of some large biomolecules. An important question arises: what range of 3D resolution is possible to achieve under experimental conditions available now. We investigated how the number and quality of diffraction images affect the 3D resolution. First the SPI experiment was simulated and reconstructed with the Dragonfly software. Then we analyzed how the number of diffraction images and the beam intensity affect the final 3D resolution. We come to the following conclusions: 1) starting from the beam intensity value (fluence) equal to 3×1012 photons / μm2 the resolution becomes to be almost constant; 2) the resolution strongly depends on the number of diffraction patterns. More than 10 000 diffraction images are required to get 4 nm resolution. [ABSTRACT FROM AUTHOR]

The traditional dictionary learning algorithm doesn't take the internal and external characteristics of the image into account in pre-process. A novel algorithm about image entropy was proposed based on gray-gradient co-occurrence matrix solution. Calculating the entropy of image patches by gray-gradient co-occurrence matrix, each image patch was classified and each type of patches was combined into training set. All kinds of sub-dictionaries were updated by singular value decomposition algorithm based on coefficient matrix. The reconstruction experiment was carried out based on the sparse representation coefficient of the test image. The simulation results show that the algorithm can effectively produce adapted sparse dictionary and significantly improve the accuracy of the reconstruction of image. [ABSTRACT FROM AUTHOR]

Compressed sampling in shift-invariant spaces (SI) is an effective method for sampling of sparse signals. But, reconstruction of compressed sampling may be unstable. In the paper, the possibility of stable reconstruction under a sufficient sparsity is proven. Further, we consider the situation where the minimal L1 norm is used to recover sparse signals from the noisy data. The result shows that they are stable. Finally, we show that the minimal L1 norm through the simulation, and explain the applicability of our algorithm to sampling systems. [ABSTRACT FROM AUTHOR]

Abstract Inpainting refers to reconstruct the incomplete image or video via analysing their context, feature of the tailing etc. Convolutional neural network with deep learning is proved to be an effective method to achieve inpainting. However, those algorithms existed now usually have vague and blurry results with huge amount of time to train the models. To address this issue, this article based on the construction of Context Encoders, continue to use the strategy of combining the encoders and generative adversarial networks (GANs), on which we add the global discriminator, consisting of the multi-scale adversarial network with the local discriminator altogether. The local discriminator ensures the local detail while the global discriminator guarantees the global consistency. Comparing with the Context Encoders, this network is simplified by reducing some of the redundant fabric, therefore this network is faster. Meantime, we re-calculate the loss function of the network and train it with the Paris dataset. The results proved that our network can achieve a better performance on street-view pictures than Context Encoders to some extent. [ABSTRACT FROM AUTHOR]

The reconstruction from Electroencephalography (EEG) signals to the corresponding image, also named RE2I, plays an important role in promoting the practical applications of brain media. However, due to the low signal-to-noise ratio (SNR) and the significant individual differences of EEG signals, extracting the underlying semantic features of EEG signals and designing a high-performance framework to implement RE2I tasks remain huge challenges. In this study, we propose a DM-RE2I framework based on the diffusion model (DM), which contains an EEG-Visual-Residual-Network (EVRNet) module and a Denoising Diffusion Probabilistic Module (DDPM). In DM-RE2I framework, we first extract the EEG semantic feature (ESF) by using the EVRNet, and then reconstruct images with the same semantics as the corresponding EEG by the DDPM. In addition, we also propose the ESF-guided DDPM (EG-DDPM) training and test algorithms for constructing matching relationship between EEG semantic features and images, as well as generating image with the same semantic as EEG semantic feature, respectively. Experimental results show DM-RE2I has a better capability of EEG semantic feature extraction, and could reconstruct the corresponding image with high resolution and high accuracy. • This paper proposes a novel EEG-based semantic feature extraction module EVRNet to extract temporal and spatial information of EEG signals. By introducing a Multi-Kernel Residual Block (MKRB), not only the gradient vanishing/exploding can be solved, but also the latent richer semantic feature in EEG can be extracted by different convolution kernels. In addition, this paper adopts an average pooling to make the EVRNet better robustness and compatibility. • This paper proposes an EEG semantic guided module named EG-DDPM, as well as the training algorithm and the test algorithm of EG-DDPM, and introduces the ConvNeXt Block into Denoising Diffusion Probabilistic Module (DDPM) Network to reconstruct the corresponding images with high resolution. • This paper designs and implements a reconstruction framework from EEG to the corresponding images, DM-RE2I, and validated its performance on two different EEG datasets and different GPUs. [ABSTRACT FROM AUTHOR]