Search

• A novel chain-shaped flow field (CSFF) is studied numerically and experimentally. • A modified agglomerate model is proposed. • The CSFF enhances under-rib convection and oxygen mass transfer. • A more uniform current density distribution is obtained in the CSFF. The flow field is crucial for the performance and efficiency of proton exchange membrane fuel cells (PEMFCs). In this research, an innovative chain-shaped flow field (CSFF) configuration is proposed and investigated, focusing on cell performance, oxygen mass transfer, and current density distribution. For this purpose, a three-dimensional multiphase PEMFC model that is based on a modified agglomerate model has been established. Simulation results demonstrate that the CSFF facilitates the under-rib convection between neighboring flow channels and boosts mass transfer from the flow channels to the catalyst layer, consequently increasing both the oxygen concentration and its uniformity within the catalyst layer. An increase of 18.9 % in the net power density of the optimized CSFF has been achieved compared to the parallel flow field (PFF). Experimental results corroborate the simulation, demonstrating that the CSFF achieves an output voltage that is 52 mV greater than the PFF at a current density of 2.9 A cm−2. Further evidence from electrochemical impedance confirms that the CSFF shows a reduced oxygen mass transfer resistance. Additionally, measurements of local current density suggest that the CSFF design enhances the uniformity of current distribution. The findings imply that the proposed CSFF shows great potential for practical application. [ABSTRACT FROM AUTHOR]

To refine the microstructure and form Ti 3 Al 2 and C14 nano particles, and then improve strength and strain simultaneously, the Ti48Al2Cr2Nb 2.5 C alloy has been prepared under ultrasonic wave. Size of lamellar colony, lamellar spacing, phase composition, compressive properties and the related mechanisms have been studied. Results show that size of lamellar colony decreases from 40.7 to 33.0 µm and lamellar spacing decreases from 1.5 to 0.8 µm. Average length-diameter ratio of Ti2AlC decreases from 5.4 to 1.9 with increasing the ultrasonic treatment time from 20 to 100 s. Two kinds of nanoparticles form. One is hexagonal Ti 3 Al 2 phase, which has incoherent relationship with γ phase. Another is Ti(Al, Cr)2 phase with C14 structure. The Ti2AlC and α grain are refined by undercooling nucleation and dendrite fragmentation in solid-liquid two-phase area. The active dislocations and vacancy under ultrasonic promote the diffusion of Ti and Cr, and then induce the Ti 3 Al 2 and Ti(Al, Cr)2 formed during solid-phase transformation. Compressive results show that the strength increases from 1901 to 2532 MPa, and the strain simultaneously increases from 27 to 34% with increasing the time from 0 to 100 s. The lattice distortion caused by the difference in thermal expansion coefficient between the Ti 3 Al 2 , Ti(Al, Cr)2 and matrix and the microstructure refined by ultrasonic jointly hinder dislocation slip to strengthen the alloy. The refinement and uniform distribution of Ti2AlC hinder crack propagation to improve toughness. [ABSTRACT FROM AUTHOR]

In this study, we used a difference-in-difference (DID) approach to analyze the effect of environmental regulation on corporate tax avoidance behavior based on China's carbon emissions trading pilot policy of 2013. Our findings were as follows: (1) Environmental regulation has led companies to adopt further tax evasion behaviors. Furthermore, the core conclusion was confirmed after a series of robust and endogenous tests, such as parallel trends and PSM-DID (propensity score matching-difference-in-difference). (2) Environmental regulations increase tax avoidance activities by reducing corporate cash flows. (3) The influence of environmental regulation on firm tax evasion is highly pronounced among non-state-owned enterprises, big-scale enterprises, and enterprises with a high degree of industry competition. [ABSTRACT FROM AUTHOR]

To achieve a more uniform distribution of the reinforcing particle and then refine them, ultrasonic treatment (UT) performs in the liquid-solid two-phase region for a certain time of Ti48Al2Cr2Nb3C alloys. The influence of uniformly distributed reinforcing particle on the microstructure evolution process is revealed through in-situ observation and the mechanism of simultaneous improvement of strength and plasticity is revealed. Results show that UT effectively eliminates the B2 phase by facilitating solute diffusion. UT successfully disperses Ti 2 AlC agglomerations and suppresses the formation of the network γ phase. The capillary and cavitation effects induced by UT contribute to the uniform distribution of Ti 2 AlC, leading to a reduction in Al enrichment between Ti 2 AlC clusters and subsequent elimination of the network γ phase. In the as-cast alloy, the presence of extensive dislocations near Ti 2 AlC is attributed to melt shrinkage during solidification. A significant number of nano-twins are observed in the vicinity of Ti 2 AlC after UT. This transformation is facilitated by the generation of Frank-Read sources induced by UT, which promote dislocation slip and twin formation. The compressive strength and strain of Ti48Al2Cr2Nb3C exhibit notable enhancements, increasing from 2088.7 MPa to 27.6% to 2422.4 MPa and 33.0%, respectively. The uniform distribution of Ti 2 AlC effectively mitigates stress concentration-induced particle breakage, while the presence of twins provides protection against dislocation-induced damage. These improvements enhance the load-bearing capacity of Ti 2 AlC, thereby improving the mechanical properties of Ti48Al2Cr2Nb3C alloy. [ABSTRACT FROM AUTHOR]

In this paper, we propose a novel time-domain end-to-end network (LaSNet) for solving the problem of multiple sound sources localization (SSL) and separation based on microphone array. The traditional time-frequency (T-F) signal representation is subject to various prior conditions and fails to separate the different sound signal components. Even the data-driven neural network does not develop an effectively integrated approach where localization and separation interplay to serve both challenge. To address the aforementioned issue, we propose a novel approach that involves the implementation of a Separation Driving Localization Network (SDLNet). This framework operates by extracting latent features from a separation network and subsequently employing them in the context of a localization network. Then we propose a simple multi-task network for both SSL and separation. Through the analysis of steering vector filter, we find that the localization and separation problems can be linked by the operation of pseudo-inverse (pinv). To facilitate a synergistic relationship between SSL and sound separation, while also enabling end-to-end network training, we develop a Pinv Module (PM). Fianlly, the Localization and Separation Network (LaSNet) structure of this paper is proposed. Inspired by the overlay mechanism of network, LasNet is extended to a multi-task and multi-layer network, in which separation task is divided into multiple subtasks. A fuzzy separation loss function is introduced for training multi-layer network. Numerical experiments demonstrate that the proposed method has a clearly better advantageous improvement than several well known models. LaSNet has greatly performance improvement in both separation and localization, and achieves at least 32% relative reduction in model size, compared with the baseline models. • The conventional multi-task network-based speech separation and source localization does not address dependency between multiple tasks. • Localization and SeparationNetwork (LaSNet) (MMNet) can be able to alleviate dependency using steering vector. • A selectable channel with residual connection can reduce computational complexity of LaSNet. • At least 32% relative reduction in model size are achieved. [ABSTRACT FROM AUTHOR]

Most wind energy harvesters (WEHs) that have been reported in the literature collect wind energy using only one type of wind-induced vibration, such as vortex-induced vibration (VIV), galloping, and flutter or wake galloping. In this letter, the interaction between VIV and galloping is used to improve the performance of WEHs. For a WEH constructed by attaching a bluff body with a rectangular cross-section to the free end of a piezoelectric cantilever, the measures to realize the interaction are theoretically discussed. Experiments verified the theoretical prediction that the WEHs with the same piezoelectric beam may demonstrate either separate or interactive VIV and galloping, depending on the geometries of the bluff bodies. For the WEHs with the interaction, the wind speed region of the VIV merges with that of the galloping to form a single region with high electrical outputs, which greatly increases the electrical outputs at low wind speeds. The interaction can be realized even when the predicted galloping critical speed is much higher than the predicted VIV critical speed. The proposed interaction is thus an effective approach to improve the scavenging efficiencies of WEHs operating at low wind speeds. [ABSTRACT FROM AUTHOR]

A deeper understanding of the internal dynamics and transient phenomena during cold start of proton exchange membrane fuel cell (PEMFC) is essential to inspire better strategies. In this study, segmented fuel cell technology is applied to capture the internal current and temperature evolution during cold start of PEMFC. It is found that the effect of flow field on cold start performance is less important than that under normal temperature. The single-channeled serpentine (SS) flow field exhibits the fastest response rate and better cold start performance owing to its high pressure drop and excellent drainage ability. The flow field does not affect the overall evolution trend, but the response rate increases with the pressure drop. During successful cold start, the maximum current density first appears in the downstream of the cathode, then it moves upwards to the middle and inlet region. However, during failed cold start, the current density gradually decreases while moving upstream due to ice blockage. Besides, severe performance degradation occurs after failed cold start. The degradation of the SS flow field is significantly greater than the others, indicating that a rapid response and superior cold start performance may increase the chance of degradation once the cold start fails. • Flow field's effect on cold start is less significant than at normal temperature. • SS flow field exhibits the fastest response and the best cold start performance. • The internal current evolution trend is not affected by the flow field structure. • Flow field with fast dynamic response aggravates degradation once cold start fails. [ABSTRACT FROM AUTHOR]

[Display omitted] • Degradation is analyzed in combination with the dynamic response during cold start. • In-plane distribution of ice during failed cold start is not uniform. • Degradation in the downstream of the cathode is more severe than in other regions. • Crack formation, Pt particle growth, and ionomer agglomeration are the root causes. Improving the durability of cold start is essential for commercializing proton exchange membrane fuel cells (PEMFCs). This study focuses on the degradation mechanism of cold start. For the first time, through applying segmented print circuit board technology and local characterizations, the degradation behavior of cold start is analyzed in combination with the dynamic response during cold start. The degradation mechanism is built from macroscopic to microscopic. The results show that cell performance and uniformity of the current density distribution are severely degraded after 30 cold start failures. Local current density tests exhibit that during failed cold start, the current is first generated downstream of the cathode. Most of the current is generated before the failing of the cold start, suggesting non-uniform in-plane ice distribution in the catalyst layer. Local electrochemical and physical characterization in the downstream, middle and upstream regions demonstrate the most serious degradation in the downstream region. In contrast, there is a slight degradation in the middle and upstream regions, illustrating the non-uniform degradation distribution during failed cold start. Meanwhile, local physical characterization indicates that freezing-induced crack formation, Pt particle growth, and ionomer agglomeration in the catalyst layer are the root causes of the freezing damage. Through comprehensive characterization, the degradation mechanism of cold start is investigated more profoundly, which is conducive to understanding the cold start process. [ABSTRACT FROM AUTHOR]

To reveal homologous substitution and refined mechanisms, and improve mechanical properties, Ti42Al6Nb 2.5 C x Ta alloys were prepared by arc melting. Element distribution, lamellar colony size, length-diameter radio of Ti 2 AlC phase, tensile properties, and the corresponding influence mechanism were studied. Results show that the relative contents of γ and Ti 2 AlC phase increase, the content of α 2 phase decreases, when Ta content increases from 0 to 2.0 at%. The relative content of B2 phase increases among the lamellar colonies and the length-diameter ratio of Ti 2 AlC particle decreases. Lamellar colony size decreases from 24.9 to 10.6 µm. With the increase of Ta content, the supercooling at the front of the solid-liquid interface increases and the β phase is refined. The increase of the precipitated phase content leads to the increase of the number of α phase heterogeneous nucleation particles, which will refine the lamellar colony. Tensile results show that the strength increases from 238 to 513 MPa and strain increases from 2.7 % to 11.1 % of Ti42Al6Nb 2.5 C1.6Ta alloy when the tested temperature increases from 750 to 950 ℃. The high temperature tensile properties of the alloy with more Ta content are relatively higher. Ta forms large lattice distortion after solid solution into the matrix due to its large atomic size, and Ta makes more Nb dissolve into the collective to increase the effect of solid solution strengthening. Another reason is the refinement of microstructure. • Alloys with finer microstructure and better high temperature tensile properties were prepared by adding different contents of Ta. • The lamellar colony and the Ti2AlC precipitates are refined, the size of lamellar colony decreases from 24.9 to 10.6 µm. • The tensile strength and elongation of alloy increase with the increase of Ta content. • Tensile results show that the strength increases from 238 to 513 MPa and strain increase from 2.7 % to 11.1 % of Ti42Al6Nb2.5C1.6Ta alloy when the tested temperature increase from 750 to 950 ℃. [ABSTRACT FROM AUTHOR]

Recently, vision-based methods have exhibited promising prospects in assigning autonomous robots more capabilities for better environment perception. Since visual sensors are easily affected under extreme conditions, current image inpainting methods based on CNNs jointly with generative adversarial networks (GANs) usually generate patches quite different from the ground truth (GT), which is harmful to autonomous robots. In this paper, we propose a novel multiscale feature alignment module with an early fusion strategy to align the left and right feature maps to better capture the motion cues between them. Then, the aligned features are fused to fill holes in the left image. To aggregate the multiscale feature maps dynamically, we propose a multiscale feature aggregation module based on an attention mechanism, of which the fusion module is designed as a symmetrical architecture to adaptively incorporate the complementary contextual correlations from different feature branches. In addition, a spatial attention module able to capture the correlations among pixels is introduced into our network to enhance the inpainting capacity and generate more refined details. To evaluate the effectiveness of our proposed method, many experiments are conducted on a stereo image dataset. The quantitative and qualitative results show that our method significantly outperforms the recent state-of-the-art image inpainting methods while running over 22 fps on a single NVIDIA RTX2080Ti GPU. • A new image inpainting method which utilizes additional image as constraints to assist the inpainting process is proposed. • The multi-scale feature alignment module based on deformable convolution can realize the feature alignment of different images. • Implicitly detect and restore damaged regions by exploiting attention maps. • Image textures and features are refined through a long-range spatial attention mechanism. [ABSTRACT FROM AUTHOR]

Starting the proton exchange membrane fuel cell (PEMFC) stack from subfreezing temperatures is a remaining challenge for the commercialization of PEMFC vehicles. This paper aims to provide an efficient cold start strategy for PEMFC stacks to meet the requirements of rapid, safe and efficient startup in a wide range of subfreezing temperatures. Herein, an efficient cold start strategy combines the advantages of self-starting and coolant heating is proposed and validated in terms of startup time, voltage consistency and energy consumption. Results demonstrate that self-starting from −20 °C is achieved within 30 s. Rapid and safe cold-start from −30 °C and −40 °C is also realized within 82 s and 100 s, respectively. The target heating temperature of −20 °C is optimal for this strategy. The voltage consistency of the stack during cold start is closely related to cold start performance, and is affected by the starting temperature and the heating power. The stack voltage consistency increases with the increase of starting temperature and heating power. Energy conservation calculations indicate that the heating power should match the stack temperature to avoid excessive energy consumption and insufficient heating. This paper provides an efficient solution for cold starting of a hybrid PEMFC system. • The metal bipolar plate PEMFC stack realized self-starting from −20 °C within 30 s. • This strategy combines the advantages of self-cold start and assisted cold start. • This strategy can achieve a rapid, safe and efficient cold start down to −40 °C. • Voltage consistency of the stack during cold start was analyzed. [ABSTRACT FROM AUTHOR]

Sketch retrieval aims at retrieving the most similar sketches from a large database based on one hand-drawn query. Successful retrieval hinges on an effective representation of sketch images and an efficient search method. In this paper, we propose a representation scheme which takes sketch strokes into account with local features, thereby facilitating efficient retrieval with codebooks. Stroke features are detected via densely sampled points on stroke lines with crucial corners as anchor points, from which local gradients are enhanced and described by a quantized histogram of gradients. A codebook is organized in a hierarchical vocabulary tree, which maintains structural information of visual words and enables efficient retrieval in sub-linear time. Experimental results on three data sets demonstrate the merits of the proposed algorithm for effective and efficient sketch retrieval. [ABSTRACT FROM AUTHOR]

We develop a framework for estimating the noise level of a natural image using two important statistics: 1) high kurtosis and 2) scale invariance in transform domain. By exploring the said priors of natural image statistics in 2-D discrete cosine transform (DCT) domain, we reveal the limitations of these statistics for images with highly directional edges or large smooth areas. Then, we derive a novel two-step estimation scheme for noise variance: 1) in preliminary estimation, an integration of wavelet and nondirectional DCT transform is used to alleviate the influence of image’s structures and 2) a noise-injection rectification is further devised to deal with the noise-free image contents. A simulation and comparative study demonstrates that this algorithm reliably infers noise variance and its robustness over wide ranges of visual content and noise levels, while outperforming some relevant methods. This paper can significantly improve the performance of existing denoising techniques that require the noise variance as a critical parameter. [ABSTRACT FROM PUBLISHER]

Natural image is characterized by its highly kurtotic and heavy-tailed distribution in wavelet domain. These typical non-Gaussian statistics are commonly described by generalized Gaussian density (GGD) or α -stable distribution. However, each of the two models has its own deficiency to capture the variety and complexity of real world scenes. Considering the statistical properties of GGD and α -stable distributions respectively, in this paper we propose a hybrid statistical model of natural image’s wavelet coefficients which is better in describing the leptokurtosis and heavy tails simultaneously. Based on a clever fusion of GGD and α -stable functions, we establish the optimal parametric hybrid model, and a close-formed Kullback–Leibler divergence of the hybrid model is derived for evaluating model accuracy. Experiment results and comparative studies demonstrate that the proposed hybrid model is closer to the true distribution of natural image’s wavelet coefficients than the single modeling using GGD or α -stable, while is beneficial for applications such as image comparison. [ABSTRACT FROM AUTHOR]

Signer adaptation is important for sign language recognition systems because a fixed system cannot perform well on all kinds of signers. In supervised signer adaptation, the labeled adaptation data must be collected explicitly. To skip the data collecting process in signer adaptation, we propose a novel unsupervised adaptation method, namely the hypothesis comparison guided cross validation method. The method not only addresses the problem of the overlap between the data set to be labeled and the data set for adaptation, but also employs an additional hypothesis comparison step to decrease the noise rate of the adaptation data set. We also utilize linguistic prior knowledge to down sample the adaptation data list to further decrease the noise rate. To evaluate the effectiveness of the proposed method, the CASIIE-SL-Database is formed, which is the first specialized data set for unsupervised signer adaptation to the best of our knowledge. Experimental results show that the proposed method can achieve relative word error rate reductions of 3.93% and 4.05% respectively compared with self-teaching method and cross validation method. Though the method is proposed for signer adaptation, it can also be applied to speaker adaptation and writer adaptation directly. [ABSTRACT FROM AUTHOR]

Quality assessment is of central importance in numerous image processing tasks. State-of-the-art objective image quality assessment (IQA) algorithms are generally devised for specific distortion types or based on training procedure of large databases. In this work, we propose a general-purpose full-reference/no-reference (FR/NR) IQA framework for image distortions, nominated by Image Quality/Distortion Metric (IQDM). The leptokurtic and heavy-tailed behaviors of image wavelet coefficients are characterized by symmetric α -stable (S α S) density, and the statistical studies indicate that the model parameters may be altered because of the presence of distortion. This important priori knowledge of original image’s distribution is then used to gauge the distortion between degraded and reference S α S models in multi-scale wavelet sub-bands. We investigate the relationship between original and degraded parameters over scales, accordingly infer the original parameters from the degraded ones. A characteristic probability density function for S α S and its closed-form Kullback–Leibler distance are derived for FR/NR-IQDM using the model parameters. Extensive experiments and comparisons demonstrate that the proposed FR/NR-IQDM scheme is efficacious to most common types of distortion, and leads to a highly comparable performance to the benchmarks and prevalent competitors in consistency with subjective judgements. [ABSTRACT FROM AUTHOR]

The analysis of user search goals for a query can be very useful in improving search engine relevance and user experience. Although the research on inferring user goals or intents for text search has received much attention, little has been proposed for image search. In this paper, we propose to leverage click session information, which indicates high correlations among the clicked images in a session in user click-through logs, and combine it with the clicked images' visual information for inferring user image-search goals. Since the click session information can serve as past users' implicit guidance for clustering the images, more precise user search goals can be obtained. Two strategies are proposed to combine image visual information with the click session information. Furthermore, a classification risk based approach is also proposed for automatically selecting the optimal number of search goals for a query. Experimental results based on a popular commercial search engine demonstrate the effectiveness of the proposed method. [ABSTRACT FROM PUBLISHER]

Abstract: In this paper, we present a generalized Earth Mover’s Distance (EMD) for pedestrian identification with body prior. (i) The general configuration of body is a valuable cue for human identification. A model of body prior about the general configuration is pursued to describe body prior for every pedestrian. In addition, the spatial incompatibility is computed by Kullback–Leibler divergence according to the pursued model and embedded into the ground distance of EMD. (ii) Furthermore, we generalize EMD by assigning different weights to regions of body, which are learned based on maximum margin criterion to boost discriminative power for pedestrian identification. The experimental results show that the generalized EMD and body prior substantially improve performance on pedestrian identification and the proposed approach is comparable to the state-of-the-art performance. [Copyright &y& Elsevier]

Aliasing is a common artifact in low-resolution (LR) images generated by a downsampling process. Recovering the original high-resolution image from its LR counterpart while at the same time removing the aliasing artifacts is a challenging image interpolation problem. Since a natural image normally contains redundant similar patches, the values of missing pixels can be available at texture-relevant LR pixels. Based on this, we propose an iterative multiscale semilocal interpolation method that can effectively address the aliasing problem. The proposed method estimates each missing pixel from a set of texture-relevant semilocal LR pixels with the texture similarity iteratively measured from a sequence of patches of varying sizes. Specifically, in each iteration, top texture-relevant LR pixels are used to construct a data fidelity term in a maximum a posteriori estimation, and a bilateral total variation is used as the regularization term. Experimental results compared with existing interpolation methods demonstrate that our method can not only substantially alleviate the aliasing problem but also produce better results across a wide range of scenes both in terms of quantitative evaluation and subjective visual quality. [ABSTRACT FROM PUBLISHER]

A set of magnetic shaping coils is installed on the Prairie View (PV) rotamak for the study of active plasma shape control in the regimes with and without toroidal field (TF). In the spherical tokamak regime (with TF), plasma current Ip can be boosted by 200% when all five shaping coils (connected in series) are energized. The enhancement of current drive efficiency is mainly attributed to the radial compression and the substantially axial extension of the plasma column; this in turn improves the impedance matching and thus increases antenna input power. In the field-reversed configuration (without TF), plasma current can be boosted by 100% when one middle coil is used; the appearance of radial shift mode limits the achievable value of Ip. The experiments clearly demonstrate that the plasma shape control plays a role in effectively driving plasma current in rotamaks. [ABSTRACT FROM AUTHOR]

Abstract: In this article, a novel algorithm – CamShift guided particle filter (CAMSGPF) – is proposed for tracking object in video sequence. CamShift is incorporated into the probabilistic framework of particle filter as an optimization scheme for proposal distribution. Meanwhile, in the context of particle filter, the scale adaptation of CamShift is improved and the computation complexity is reduced. It is demonstrated through several real tracking tasks that the new method performs better than baseline trackers in both tracking robustness and computational efficiency. [Copyright &y& Elsevier]

A rapid and safe cold start is essential for the commercialisation of proton exchange membrane fuel cell (PEMFC) vehicles, and self-start is considered as the optimal way to fulfil this expectation. In this study, the self-start of large-area PEMFCs is investigated. It is found that there exists a severe temperature distribution nonuniformity along the flow direction during one-way air-supply self-start, and the cold-start capability is limited by the fact that most of the heat is concentrated in the inlet area, while less heat is in the outlet area, which makes it difficult to start at −15 °C. To solve this issue, an alternating air-supply strategy is suggested. By applying this strategy, temperature distribution uniformity and cold-start performance are significantly improved. Self-start from −15 °C and −20 °C can be achieved in 165 s and 142 s with starting voltages of 0.8 V and 0.6 V, respectively. The alternating interval has a negligible impact on start-up time but will affect the temperature distribution and fluctuation. The advantages of alternate air supply are analyzed based on the calculation of heat generation and water production. A performance degradation test indicates that cell performance is well preserved after 20 cold starts [Display omitted] [ABSTRACT FROM AUTHOR]

A study of magnetic field structure evolution during 40-ms plasma discharge has been performed in a new device with 80 cm long/40 cm diameter cylindrical chamber, in which a plasma current Ip≈2 kA was driven and sustained by a rotating magnetic field. The main focus of the experiments is on how the changes in externally applied magnetic field affect the current profile and magnetic field in plasma. During plasma discharge, a pulse current was briefly fed to a magnetic coil located at the midplane (middle coil). The magnetic field in cross section of plasma was scanned with pickup probes. Two regimes were studied: without and with an external toroidal field (TF) produced by axial Iz current. With a relatively small current (Im ≲ 600 A) in the middle coil, the plasma current is boosted up to 5 kA. The magnetic flux surfaces become extended along the axial Z direction, sometimes with the formation of doublet shape plasma. The regime without TF appears to be less stable, presumably due to the reversal of plasma current in central area of plasma column. [ABSTRACT FROM AUTHOR]

Abstract: This paper firstly presents a study of video quality for H.264 compressed videos compared to MPEG-4 (simple profile) from a perceptual point of view. Traditionally, peak signal-to-noise ratio (PSNR) has been used to represent the quality of a compressed video sequence. However, PSNR has been found to correlate poorly with subjective quality ratings, particularly at much lower bitrates and low frame rates. Thus, an alternative that has been commonly used is to perform subjective test where a large number of human subjects are used to gauge the quality of a video. However, this process is not only time-consuming but also tedious and expensive to perform. Hence, this paper further proposes an objective video quality measurement method to automatically measure the perceptual quality of a stream of video images. The proposed method has been tested on multimedia videos, consisting of CIF and QCIF video sequences compressed at various bitrates (24–384kbps) and frame rates (7.5–30fps) using the H.264 and MPEG-4 video codecs and it is shown to give significantly better correlations to the human perception than PSNR and the video structural similarity method. [Copyright &y& Elsevier]

A highly efficient quasi-optical mode converter system with several novel features has been designed and tested at Forschungszentrum Karlsruhe (FZK). The converter consists of a dimpled-wall waveguide launcher, one quasi-elliptical mirror and two toroidal mirrors. The coupled-mode theory has been used to analyze the operation of the prebunching waveguide launcher; the radiated fields from the cut of the launcher have been calculated by the scalar diffraction integral. Simulation results show that the advanced dimpled-wall launcher generates a well-focused Gaussian radiation pattern with low diffraction losses. In this case, toroidal mirrors are sufficient to obtain a desired output beam pattern. An efficiency of more than 98% has been achieved to convert the rotating TE28,8 cavity mode at 140 GHz into a fundamental Gaussian beam. Experimental measurements show close agreement with theoretical predictions. [ABSTRACT FROM AUTHOR]

Membrane water content is of vital importance to the freezing durability of proton exchange membrane fuel cells (PEMFCs). Excessive water freezing could cause irreversible degradation to the cell components and deteriorate the cell performance and lifetime. However, there are few studies on the critical membrane water content, a threshold beyond which freezing damage occurs, for cold storage of PEMFCs. In this work, we first proposed a method for measuring membrane water content using membrane resistance extracted from measured high frequency resistance (HFR) based on the finding that the non-membrane resistance part of the measured HFR is constant within the range of membrane water content of 2.98 to 14.0. Then, freeze/thaw cycles were performed from −50 °C to 30 °C with well controlled membrane water content. After 30 cycles, cells with a membrane water content of 8.2 and 7.7 exhibited no performance degradation, while those higher than 8.2 showed significant performance decay. Electrochemical tests revealed that electrochemical surface area (ECSA) reduction and charge transfer resistance increase are the main reasons for the degradation. These results indicate that the critical membrane water content for successful cold storage at −50 °C is 8.2. [ABSTRACT FROM AUTHOR]

Diabetic retinopathy (DR) is a common complication of diabetes mellitus (DM), and it is necessary to diagnose DR in the early stages of treatment. With the rapid development of convolutional neural networks in the field of image processing, deep learning methods have achieved great success in the field of medical image processing. Various medical lesion detection systems have been proposed to detect fundus lesions. At present, in the image classification process of diabetic retinopathy, the fine-grained properties of the diseased image are ignored and most of the retinopathy image data sets have serious uneven distribution problems, which limits the ability of the network to predict the classification of lesions to a large extent. We propose a new non-homologous bilinear pooling convolutional neural network model and combine it with the attention mechanism to further improve the network's ability to extract specific features of the image. The experimental results show that, compared with the most popular fundus image classification models, the network model we proposed can greatly improve the prediction accuracy of the network while maintaining computational efficiency. [ABSTRACT FROM AUTHOR]

The catalytic hydrogen/oxygen reaction method is one of the primary ways to realize the rapid cold start of proton exchange membrane fuel cell (PEMFC). In this study, the start-up capacity of the catalytic hydrogen/oxygen reaction method is systematically investigated. A tremendous nonuniformity of temperature distribution is found under the condition of one-way mixture gas supply. A reverse feeding unit (RFU) is invented to realize a double-way gas supply to solve this problem. With the aid of RFU, the temperature uniformity during cold start is greatly improved. When the flow rate is 30 L min−1 and air fraction is 24%, the average and the lowest temperature take 55 s and 79 s respectively to rise from −32 °C to 0 °C, and the maximum temperature variance is 12.6. However, the lowest temperature of stack only arrives at −11.8 °C in 79 s when one-way mixture gas is supplied, and the temperature variance evidently increase to 21.4. After the preheating period, the output performance using RFU is 6.01% higher than the one-way gas supply method when the loading current is 250 mA cm−2. The temperature uniformity and power output is improved remarkably by the RFU. • Nonuniform distribution of temperature is found under the one-way gas supply condition. • The reason for the nonuniformity is uncovered by thermal calculation. • The nonuniformity is alleviated by a reverse feeding unit (RFU). • PEMFC stack starts up in 79 s with the uniform temperature distribution from −32 °C. [ABSTRACT FROM AUTHOR]

Graph matching involves combinatorial optimization based on edge-to-edge affinity matrix, which can be generally formulated as Lawler's quadratic assignment problem (QAP). This paper presents a QAP network directly learning with the affinity matrix (equivalently the association graph) whereby the matching problem is translated into a constrained vertex classification task. The association graph is learned by an embedding network for vertex classification, followed by Sinkhorn normalization and a cross-entropy loss for end-to-end learning. We further improve the embedding model on association graph by introducing Sinkhorn based matching-aware constraint, as well as dummy nodes to deal with unequal sizes of graphs. To our best knowledge, this is one of the first network to directly learn with the general Lawler's QAP. In contrast, recent deep matching methods focus on the learning of node/edge features in two graphs respectively. We also show how to extend our network to hypergraph matching, and matching of multiple graphs. Experimental results on both synthetic graphs and real-world images show its effectiveness. For pure QAP tasks on synthetic data and QAPLIB benchmark, our method can perform competitively and even surpass state-of-the-art graph matching and QAP solvers with notable less time cost. We provide a project homepage at http://thinklab.sjtu.edu.cn/project/NGM/index.html. [ABSTRACT FROM AUTHOR]

Motivation: Retrosynthesis planning poses a formidable challenge in the organic chemical industry, particularly in pharmaceuticals. Single-step retrosynthesis prediction, a crucial step in the planning process, has witnessed a surge in interest in recent years due to advancements in AI for science. Various deep learning-based methods have been proposed for this task in recent years, incorporating diverse levels of additional chemical knowledge dependency. Results: This paper introduces UAlign, a template-free graph-to-sequence pipeline for retrosynthesis prediction. By combining graph neural networks and Transformers, our method can more effectively leverage the inherent graph structure of molecules. Based on the fact that the majority of molecule structures remain unchanged during a chemical reaction, we propose a simple yet effective SMILES alignment technique to facilitate the reuse of unchanged structures for reactant generation. Extensive experiments show that our method substantially outperforms state-of-the-art template-free and semi-template-based approaches. Importantly, our template-free method achieves effectiveness comparable to, or even surpasses, established powerful template-based methods. Scientific contribution: We present a novel graph-to-sequence template-free retrosynthesis prediction pipeline that overcomes the limitations of Transformer-based methods in molecular representation learning and insufficient utilization of chemical information. We propose an unsupervised learning mechanism for establishing product-atom correspondence with reactant SMILES tokens, achieving even better results than supervised SMILES alignment methods. Extensive experiments demonstrate that UAlign significantly outperforms state-of-the-art template-free methods and rivals or surpasses template-based approaches, with up to 5% (top-5) and 5.4% (top-10) increased accuracy over the strongest baseline. [ABSTRACT FROM AUTHOR]

Mechanical stretching can accelerate Form II to Form I phase transitions in isotatic polybutene-1 (iPB-1). This study in-situ investigated the effect of pre-stretching on further phase transitions. Three tensile modes were considered. In mode A, the samples were stretched to their breaking point, and the phase transition was wholly completed. In mode B, the samples were pre-stretched to specific strains and annealed under quiescent conditions at room temperature. Mode C was studied under the same conditions as Mode B, but the samples were allowed free shrinkage during the phase transition. The phase transition process were investigated by mechanical testing and in situ wide-angle X-ray diffraction. The phase transition kinetics were obviously faster in the pre-stretched samples (even when stretched in the elastic zone) than in the un-stretched samples. The pre-stretching may affect the conformation or the lattice arrangement of the segments in Form II, thereby accelerating the subsequent phase transition. In the samples with pre-strain beyond the yield strain, the accelerated transition rate increased with the increased pre-strain, meanwhile, the growth dimension of Form I reduced, owing to the confinement effect of stretch-induced Form I. The residual stress in the mode-B stretched samples did not significantly affect the phase transition, because the residual stress was lower than the critical stress that triggers Form I nucleation. In summary, a small pre-strain accelerated the phase transition and affected the growth dimension of Form I. Image 1 • The phase transition from form II to form I in iPB-1 under three tensile modes was studied by wide angle X-ray diffraction technique. • The phase transition kinetics of the pre-stretched samples was obviously accelerated, even at very small strain in the elastic zone. • In samples with pre-strain beyond the yield strain, the growth dimension of Form I decreased due to the confinement effect. [ABSTRACT FROM AUTHOR]

Spontaneously phase transition from form II to I in isotactic polybutene-1 (iPB-1) products during storage at room temperature after processing limited their application. Investigations on phase transition in iPB-1 have practical importance, especially on oriented iPB-1 as oriented form II generated during processing where intense flow field is always imposed. The effects of flow on phase transition were studied with synchrotron radiation scanning X-ray microdiffraction (SR- μ SXRD) technique with beam size of 4.5 × 5.2 μm2 in the present work. Localized flow-field was imposed by pulling glass fiber through undercooled melt. Benefiting from the high resolution of SR- μ SXRD technique, the formation of fibrillar and oriented form II ( c I I f and c I I o ) were identified. The transformation from c I I f to fibrillar form I ( c I f ) completed in about 40 min, while c I I o transformed into oriented form I ( c I o ) partially in about 186 min. Quantitative analysis on SR- μ SXRD result showed that crystallographic lattice dimension of c I I f changed comparing with that of c I I o , where a and b increased while c decreased in c I I f . An obvious correlation between crystallographic lattice deformation and phase transition kinetics was is identified. Local internal stress plays an important role in promoting the phase transition kinetics, as the stress transmission efficiency in c I I f with rod-like morphology would be higher than that in c I I o with lamellar morphology. Image 1 • With in-situ SR- μ SXRD measurement, it's found that fibrillar and oriented form II were induced at high and low shear rate, respectively. • Fibrillar form II transformed into form I completely in about 40 min, which is much faster than that of oriented form II. • Deformation in lattice dimension of fibrillar form II is observed which may be due to partially conformation transition before lattice rearrangement. [ABSTRACT FROM AUTHOR]

Temporal point process (TPP) is an expressive tool for modeling the temporal pattern of event sequences. However, discovering temporal patterns for event sequences clustering is rarely studied in TPP modeling. To solve this problem, we take a reinforcement learning view whereby the observed sequences are assumed to be generated from a mixture of latent policies. The purpose is to cluster the sequences with different temporal patterns into the underlying policies while learning each of the policy model. The flexibility of our model lies in: i) all the components are networks including the policy network for modeling the temporal point process; ii) to handle varying-length event sequences, we resort to inverse reinforcement learning by decomposing the observed sequence into states (RNN hidden embedding of history) and actions (time interval to next event) in order to learn a reward function, it helps to achieve better performance or increasing efficiency compared to existing methods using rewards over the entire sequence such as log-likelihood or Wasserstein distance. We adopt an Expectation-Maximization algorithm, in E-step estimating the cluster labels for each sequence, in M-step aiming to learn the respective policy. Extensive experiments on synthetic and real-world datasets show the efficacy of our method against the state-of-the-arts. [ABSTRACT FROM AUTHOR]

To form in situ nano-borides in the solid-liquid two-phase region, Ti46 A l8Nb2·6C0.8Ta x B alloys were prepared under ultrasound to improve the microstructural and mechanical properties. The morphology and distribution of borides and the characteristics of lamellar colonies were observed with scanning and transmission electron microscopy and X-ray diffraction; these analyses revealed the influence of borides on microstructural evolution and improved strength and plasticity. The borides that form are elongated and spherical. As the B content increases, the borides gradually spheroidise, and nano-borides precipitate on the lamellar phase. There are a large number of dislocations around the borides to form stacking faults and dislocation walls. The average size of the lamellar colonies decreases from 26.9 to 19.3 μm. The increase in boride nucleation sites as the B content increases and ultrasound-mediate breaking of primary borides are the reasons for forming B27-type borides. The increase in nucleation particles refines the lamellar colony. The compressive strength increases from 2058 to 2299 MPa, and the strain increases from 25.2% to 32.3% when the addition of B increases from 0% to 1.2%. The significantly improved mechanical properties are due to uniformly distributed nano-borides that hinder the movement of dislocations to form stacking faults and dislocation walls and refinement of the lamellar colonies. [ABSTRACT FROM AUTHOR]

Video prediction is challenging as real-world motion dynamics are usually multi-modally distributed. Existing stochastic methods commonly formulate random noise input with simple prior distribution, which is insufficient to model highly complex motion dynamics. This work proposes a progressive multiple granularity analysis framework to tackle the above difficulty. Firstly, to achieve coarse alignment, the input sequence is matched to prototype motion dynamics in the training set, based on self-supervised auto-encoder learning via motion/appearance disentanglement. Secondly, motion dynamics is transferred from the matched prototype sequence to input sequence via adaptively learned kernel, and the predicted frames are further refined through a motion-aware prediction model. Extensive qualitative and quantitative experiments on three widely used video prediction datasets demonstrate that: (1) the proposed framework essentially decomposes the hard task into a series of more approachable sub-tasks where a better solution is easier to be sought and (2) our proposed method performs favorably against state-of-the-art prediction methods. [ABSTRACT FROM AUTHOR]

Stereoscopic 3D (S3D) geometric distortions can be introduced by mismatches among image capture, display, and viewing configurations. In previous work of S3D geometric models, geometric distortions have been analyzed from a third-person perspective based on the binocular depth cue (i.e., binocular disparity). A third-person perspective is different from what the viewer sees since monocular depth cues (e.g., linear perspective, occlusion, and shadows) from different perspectives are different. However, depth perception in a 3D space involves both monocular and binocular depth cues. Geometric distortions that are solely predicted by the binocular depth cue cannot describe what a viewer really perceives. In this paper, we combine geometric models and retinal disparity models to analyze geometric distortions from the viewer's perspective where both monocular and binocular depth cues are considered. Results show that binocular and monocular depth-cue conflicts in a geometrically distorted S3D space. Moreover, user-initiated head translations averting from the optimal viewing position in conventional S3D displays can also introduce geometric distortions, which are inconsistent with our natural 3D viewing condition. The inconsistency of depth cues in a dynamic scene may be a source of visually induced motions sickness. [ABSTRACT FROM AUTHOR]

Video prediction refers to predicting and generating future video frames given a set of consecutive frames. Conventional video prediction methods usually criticize the discrepancy between the ground-truth and predictions frame by frame. As the prediction error accumulates recursively, these methods would easily become out of control and are often confined to the short-term horizon. In this paper, we introduce a retrospection process to rectify the prediction errors beyond criticizing the future prediction. The introduced retrospection process is designed to look back what have been learned from the past and rectify the prediction deficiencies. To this end, we build a retrospection network to reconstruct the past frames given the currently predicted frames. A retrospection loss is introduced to push the retrospection frames being consistent with the observed frames, so that the prediction error is alleviated. On the other hand, an auxiliary route is built by reversing the flow of time and executing a similar retrospection. These two routes interact with each other to boost the performance of retrospection network and enhance the understanding of dynamics across frames, especially for the long-term horizon. An adversarial loss is employed to generate more realistic results in both prediction and retrospection process. In addition, the proposed method can be used to extend many state-of-the-art video prediction methods. Extensive experiments on the natural video dataset demonstrate the advantage of introducing the retrospection process for long-term video prediction. [ABSTRACT FROM AUTHOR]

Visual tracking is challenging as target objects often undergo significant appearance changes caused by deformation, abrupt motion, background clutter and occlusion. In this paper, we propose to exploit the rich hierarchical features of deep convolutional neural networks to improve the accuracy and robustness of visual tracking. Deep neural networks trained on object recognition datasets consist of multiple convolutional layers. These layers encode target appearance with different levels of abstraction. For example, the outputs of the last convolutional layers encode the semantic information of targets and such representations are invariant to significant appearance variations. However, their spatial resolutions are too coarse to precisely localize the target. In contrast, features from earlier convolutional layers provide more precise localization but are less invariant to appearance changes. We interpret the hierarchical features of convolutional layers as a nonlinear counterpart of an image pyramid representation and explicitly exploit these multiple levels of abstraction to represent target objects. Specifically, we learn adaptive correlation filters on the outputs from each convolutional layer to encode the target appearance. We infer the maximum response of each layer to locate targets in a coarse-to-fine manner. To further handle the issues with scale estimation and re-detecting target objects from tracking failures caused by heavy occlusion or out-of-the-view movement, we conservatively learn another correlation filter, that maintains a long-term memory of target appearance, as a discriminative classifier. We apply the classifier to two types of object proposals: (1) proposals with a small step size and tightly around the estimated location for scale estimation; and (2) proposals with large step size and across the whole image for target re-detection. Extensive experimental results on large-scale benchmark datasets show that the proposed algorithm performs favorably against the state-of-the-art tracking methods. [ABSTRACT FROM AUTHOR]

Learning to predict future visual dynamics given input video sequences is a challenging but essential task. Although many stochastic video prediction models are proposed, they still suffer from "multi-modal entanglement", which refers to the ambiguity of learned representations for multi-modal dynamics modeling. While most existing video prediction models are label-free, we propose a self-supervised labeling strategy to improve spatiotemporal prediction networks without extra supervision. Starting from a set of clustered pseudo-labels, our framework alternates between model optimization and label updating. The key insight of our method lies in that we exploit the reconstruction error from the optimized model itself as an indicator to progressively refine the label assignment on the training set. The two steps are interdependent, with the predictive model guiding the direction of label updates, and in turn, effective pseudo-labels further help the model learn better disentangled multi-modal representation. Experiments on two different video prediction datasets demonstrate the effectiveness of the proposed method. • A deep analysis of the multi-modal entanglement issue in video prediction task is provided. • A novel plug-and-play self-labeling framework is proposed for video prediction. • The proposed framework provides significant improvements on different video prediction models on two different benchmarks. [ABSTRACT FROM AUTHOR]

Style transfer describes the rendering of an image’s semantic content as different artistic styles. Recently, generative adversarial networks (GANs) have emerged as an effective approach in style transfer by adversarially training the generator to synthesize convincing counterfeits. However, traditional GAN suffers from the mode collapse issue, resulting in unstable training and making style transfer quality difficult to guarantee. In addition, the GAN generator is only compatible with one style, so a series of GANs must be trained to provide users with choices to transfer more than one kind of style. In this paper, we focus on tackling these challenges and limitations to improve style transfer. We propose adversarial gated networks (Gated-GAN) to transfer multiple styles in a single model. The generative networks have three modules: an encoder, a gated transformer, and a decoder. Different styles can be achieved by passing input images through different branches of the gated transformer. To stabilize training, the encoder and decoder are combined as an auto-encoder to reconstruct the input images. The discriminative networks are used to distinguish whether the input image is a stylized or genuine image. An auxiliary classifier is used to recognize the style categories of transferred images, thereby helping the generative networks generate images in multiple styles. In addition, Gated-GAN makes it possible to explore a new style by investigating styles learned from artists or genres. Our extensive experiments demonstrate the stability and effectiveness of the proposed model for multi-style transfer. [ABSTRACT FROM AUTHOR]

Person search aims to simultaneously localize and identify a query person from uncropped images. To achieve this goal, state-of-the-art models typically add a re-id branch upon two-stage detectors like Faster R-CNN. Owing to the ROI-Align operation, this pipeline yields promising accuracy as re-id features are explicitly aligned with the corresponding object regions, but in the meantime, it introduces high computational overhead due to dense object anchors. In this work, we present an anchor-free approach to efficiently tackling this challenging task, by introducing the following dedicated designs. First, we select an anchor-free detector (i.e., FCOS) as the prototype of our framework. Due to the lack of dense object anchors, it exhibits significantly higher efficiency compared with existing person search models. Second, when directly accommodating this anchor-free detector for person search, there exist several misalignment issues in different levels (i.e., scale, region, and task). To address these issues, we propose an aligned feature aggregation module to generate more discriminative and robust feature embeddings. Accordingly, we name our framework as Feature-Aligned Person Search Network (AlignPS). Third, by investigating the advantages of both anchor-based and anchor-free models, we further augment AlignPS with an ROI-Align head, which significantly improves the robustness of re-id features while still keeping our model highly efficient. Our framework not only achieves state-of-the-art or competitive performance on two challenging person search benchmarks, but can be also extended to other challenging searching tasks such as animal and object search. All the source codes, data, and trained models are available at: https://github.com/daodaofr/alignps. [ABSTRACT FROM AUTHOR]

Object tracking is challenging as target objects often undergo drastic appearance changes over time. Recently, adaptive correlation filters have been successfully applied to object tracking. However, tracking algorithms relying on highly adaptive correlation filters are prone to drift due to noisy updates. Moreover, as these algorithms do not maintain long-term memory of target appearance, they cannot recover from tracking failures caused by heavy occlusion or target disappearance in the camera view. In this paper, we propose to learn multiple adaptive correlation filters with both long-term and short-term memory of target appearance for robust object tracking. First, we learn a kernelized correlation filter with an aggressive learning rate for locating target objects precisely. We take into account the appropriate size of surrounding context and the feature representations. Second, we learn a correlation filter over a feature pyramid centered at the estimated target position for predicting scale changes. Third, we learn a complementary correlation filter with a conservative learning rate to maintain long-term memory of target appearance. We use the output responses of this long-term filter to determine if tracking failure occurs. In the case of tracking failures, we apply an incrementally learned detector to recover the target position in a sliding window fashion. Extensive experimental results on large-scale benchmark datasets demonstrate that the proposed algorithm performs favorably against the state-of-the-art methods in terms of efficiency, accuracy, and robustness. [ABSTRACT FROM AUTHOR]

Information security, the practice to protect information from unauthorized use, attracts researchers attention. In this paper, we proposed a visual masking effect based perceptual information protection scheme for display devices, such as mobile device, personal computer and ATM machine. The private information is embedded into the misleading image with the proposed algorithm, resulting in a dramatic descent in perceptual saliency of the private information for peepers viewing with naked eyes, while maintaining accessibility for authorized viewers wearing color filter. Quantitative and qualitative experiments are conducted, and results show effectiveness of our algorithm. [ABSTRACT FROM AUTHOR]

A movie summarization model can automatically edit a condensed version of a movie by selecting keyframes. Some previous works have proposed some movie summarizers based on traditional methods or recent neural networks and achieved some progress. Despite the demonstrated successes, there are some limitations: (1) previous works mainly resort to hand-crafted heuristics and most of them are unsupervised; (2) currently there is no publicly suitable dataset available for the supervised movie summarization; (3) existing works only focus on the movies themselves while neglecting the audiences, who have the most to say in which part of the movie is more attractive. To break through the aforementioned limitations, we establish a movie summarization dataset Movie50 and propose a novel human attention based annotation pipeline. Furthermore, we propose the A/V-MSNet, an audiovisual neural network that takes advantage of spatio-temporal visual and auditory information to better simulate human attention as well as exploit more plentiful information. The network is designed, trained end-to-end, and evaluated on the public dataset and our dataset. Extensive experiments demonstrate the superiority of the proposed method. [ABSTRACT FROM AUTHOR]

A fast ion Doppler spectroscopy (IDS) diagnostic system is installed on the Prairie View Rotamak to measure ion temperature and plasma flow. The diagnostic employs a single channel photomultiplier tube and a Jarrell-Ash 50 monochromator with a diffraction grating line density of 1180 lines/mm, which allows for first order spectra of 200-600 nm. The motorized gear of the monochromator allows spectral resolution of 0.01 nm. Equal IDS measurements are observed for various impurity emission lines of which carbon lines exhibit stronger intensities. Furthermore, the diagnostics is examined in an experiment where plasma experiences sudden disruption and quick recovery. In this case, the IDS measurements show ∼130% increase in ion temperature. Flow measurements are shown to be consistent with plasma rotation. [ABSTRACT FROM AUTHOR]

Current saliency prediction models based on convolutional neural networks (CNNs) achieve solid improvement in predicting human attention on omnidirectional image (ODI). However, these models that employ standard convolution have two main shortcomings: content-agnostic and computation-intensive. To address these two shortcomings, we propose a decoupled dynamic group equivariant filter (DDGF). Specifically, inspired by the attention mechanism that adopts light-weight branches for estimating spatial and channel attention, we decouple group equivariant convolution (i.e. p4 convolution) into spatial and channel dynamic group equivariant filters. Such a design not only makes p4 convolution filter adaptive to ODI content, but also considerably reduces computational cost. To our best knowledge, the DDGF is the first decoupled dynamic convolution filter that applied to the task of saliency prediction. Meanwhile, we observe that it is effective and efficient when replacing standard group equivariant convolution with DDGF in ODI saliency prediction. Experimental results show that the proposed DDGF can achieve superior performance in comparison with other state-of-the-art methods. Additionally, we conduct ablation experiments to verify the effectiveness of each component of the proposed DDGF. [ABSTRACT FROM AUTHOR]

In this paper, we propose a wave leader pyramids based Visual Information Fidelity method for image quality assessment. Motivated by the observations that the human vision systems (HVS) are more sensitive to edge and contour regions and that the human visual sensitivity varies with spatial frequency, we first introduce the two-dimensional wavelet leader pyramids to robustly extract the multiscale information of edges. Based on the wavelet leader pyramids, we further propose a visual information fidelity metric to evaluate the quality of images by quantifying the information loss between the original and the distorted images. Experimental results show that our method outperforms many state-of-the-art image quality metrics. [ABSTRACT FROM AUTHOR]

In this paper, we propose a framework for image classification. An image is represented by multiple feature channels which are computed by the bag-of-words model and organized in a spatial pyramid. The main difference among feature channels resides in what type of base descriptor in the bag-of-words model is extracted. The overall features achieve different levels of the trade-off between discriminative power and invariance. Support vector machines with kernels based on histogram intersection distance and χ2 distance are used to obtain a posteriori probabilities of the image in each feature channel. Then, four data fusion strategies are proposed to combine intermediate results from multiple feature channels. Experimental results show that almost all the proposed strategies can significantly improve the classification accuracy as compared with the single cue methods and, especially, prod-max performs best in all experiments. The framework appears to be general and capable of handling diverse classification problems due to the multiple-feature-channel-based representation. Also, it is demonstrated that the proposed method achieves higher, or comparable, classification accuracies with less computational cost as compared with other multiple cue methods on challenging benchmark datasets. [ABSTRACT FROM AUTHOR]

Image interpolation addresses the problem of obtaining high resolution (HR) images from its low resolution (LR) counterparts. For observed LR images with aliasing artifacts caused by undersampling, commonly used interpolation methods cannot recover HR images well, and may often interpolate over-fitting artifacts. In this paper, based on the observation that natural images normally have redundant similar patches, a new patch-synthesis-based interpolation method is proposed for image interpolation. In the proposed method, an inference method based on Markov chain is adopted to select the best patches from the input LR image and synthesize them into the undersampled areas of a desired HR image. In order to improve the efficiency of the algorithm, we also introduce fields of experts to model the sparse prior knowledge and use it to measure the compatibilities among neighboring patches. Experimental results compared with traditional interpolation methods demonstrate that our method cannot only alleviate the aliasing artifact, but also produce better results in terms of quantitative evaluation and subjective visual quality [ABSTRACT FROM AUTHOR]