Medical image segmentation based on neural networks is pivotal in promoting digital health equity. The attention mechanism increasingly serves as a key component in modern neural networks, as it enables the network to focus on regions of interest, thus improving the segmentation accuracy in medical images. However, current attention mechanisms confront an accuracy-complexity trade-off paradox: accuracy gains demand higher computational costs, while reducing complexity sacrifices model accuracy. Such a contradiction inherently restricts the real-world deployment of attention mechanisms in resource-limited settings, thus exacerbating healthcare disparities. To overcome this dilemma, we propose a parameter-free Neighborhood Self-Dissimilarity Attention (NSDA), inspired by radiologists' diagnostic patterns of prioritizing regions exhibiting substantial differences during clinical image interpretation. Unlike pairwise-similarity-based self-attention mechanisms, NSDA constructs a size-adaptive local dissimilarity measure that quantifies element-neighborhood differences. By assigning higher attention weights to regions with larger feature differences, NSDA directs the neural network to focus on high-discrepancy regions, thus improving segmentation accuracy without adding trainable parameters directly related to computational complexity. The experimental results demonstrate the effectiveness and generalization of our method. This study presents a parameter-free attention paradigm, designed with clinical prior knowledge, to improve neural network performance for medical image analysis and contribute to digital health equity in low-resource settings. The code is available at [https://github.com/ChenJunren-Lab/Neighborhood-Self-Dissimilarity-Attention](https://github.com/ChenJunren-Lab/Neighborhood-Self-Dissimilarity-Attention).

PHP Object Injection (POI) vulnerabilities enable unexpected execution of class methods in PHP applications, resulting in various attacks. In the meanwhile, designing effective patches for POI vulnerabilities demands substantial engineering efforts. Existing research mostly focused on the detection of POI gadget chains, whereas the automatic patch generation remains an under-explored problem. In this work, we empirically study known gadget chains, and discover that adversaries usually construct gadget chains by diverging the execution to paths that developers never considered. The methods that get unexpectedly jump into (i.e., executed) are referred to as possible methods (PM). Based on the observation, we propose PFortifier, a framework for automatic POI patch generation. PFortifier operates in two stages: (i) the gadget chain detection phase, in which PFortifier simulates the execution of PHP applications, and detects gadget chains that pass attacker controlled objects to dangerous sinks, and (ii) the patch generation phase, in which PFortifier automatically generates POI patches by restricting PM jumps detected in the first phase. We evaluate PFortifier on 31 PHP applications and frameworks. The experiment results demonstrate the effectiveness of PFortifier: it generates precise patches for 52.53% of gadget chains, and suggests potential patches for 45.45% chains, resulting in a total chain coverage of 97.98%.

Abstract The privacy of irrelevant individuals appearing in social media videos is often compromised, which presents a challenging issue for privacy protection. Current approaches to privacy protection primarily rely on manually identifying irrelevant individuals, which is both time-consuming and labor-intensive. Therefore, this paper proposes an automatic algorithm for identifying irrelevant individuals to protect their privacy efficiently. The method employs multi-object tracking to extract spatiotemporal motion features, enabling the automated recognition of irrelevant individuals. In addition, a trajectory association algorithm is employed to improve the precision of tracking from occlusion and blurring during motion. Additionally, as the field of irrelevant people privacy protection lacks data support, this paper constructs a dataset for further research, consisting of 101 685 irrelevant faces and 40 247 relevant faces. Through experimental validation on various datasets, the proposed method shows significant improvements in various metrics compared with the state-of-the-art approaches, with MOTA increasing by 3.2%, HOTA increasing by 13%, and the accuracy of individual irrelevant recognition reaching 96.78%.

Abstract The spread of disinformation on online social media has caused massive concern. Existing disinformation detection methods neglect the diverse compositional forms of tweets in real-life scenarios, making them less applicable and effective in social media settings. Meanwhile, these methods use pattern cues but overlook important aspects such as syntax, lexicon, and shallow visual semantics, and lack attention to factual content such as time, place, and person relay in both text and images, thus failing to fully explore features of disinformation and limiting detection accuracy. Furthermore, with the popularity of large language models (LLMs), the tweets generated by these models make the style of disinformation more subtle. Since existing datasets are mostly human-generated and lack style diversity, it results in weak detection capabilities of methods trained on these datasets. To address these challenges, a dual-feature adaptive framework for multimodal disinformation detection is proposed. The framework first using a similarity-based algorithm adaptively handles different tweet forms. It then enhances pattern features by bridging multimodal output from single-modal pretrained modal, and factual features are subsequently extracted using a zero-shot method based on a large vision language model. Finally, an expert network aggregates and reweights the dual-feature representation for tweets using an LLM-text detector in gating strategy. This paper also presents two multimodal disinformation datasets that include both LLM-generated and human-generated tweets reflecting real-world scenarios. The true tweets in datasets are diverse in style, while the fake tweets are more misleading. Experimentally verified, the proposed method outperforms baseline methods by an accuracy of 1.04% and 0.72% on typical datasets while also achieving a minimum accuracy drop of 0.65% and 0.87% on the proposed dataset.

Abstract Image forgery localization identifies tampered regions within an image by extracting distinctive forgery features. Current methods mainly use convolutional neural networks (CNNs) to extract features. However, CNNs’ limited receptive field emphasizes local features, impeding the global modeling of crucial lower-level features like edges and textures, leading to decreased precision. Moreover, prior methods use pyramid networks for multi-scale feature extraction but show deficiencies in multi-scale and interlayer modeling, leading to inadequate multi-scale information representation and limiting flexibility to tampered regions of varying sizes. To address these issues, this paper proposes a Transformer-based model integrating multi-scale and boundary features. The model employs a Pyramid Vision Transformer as the encoder, using self-attention over convolution to enhance global context modeling. Building on this, the model incorporates a multi-scale feature enhancement module that enriches forgery features by paralleling various convolutional layers. Features at various encoder stages are integrated through a cross-stage interaction module, enabling multi-level feature correlation for a strong feature representation. Furthermore, the model includes a forgery boundary information-guided branch, which focuses precisely on tampered region structures without introducing irrelevant noise. Experiments demonstrate that our model surpasses previous methods in localization accuracy, with F1 and AUC improving by 8.5% and 2.2% in pre-training, respectively.

Given water-surface uncertainty problem, it is difficult for green lasers to detect the exact water surface. The near water surface penetration (NWSP) of green laser in water is not beneficial for high-accuracy water depth measurements but provides a possible mean for water–land discrimination. A novel water–land discriminator based on the NWSP of green laser is proposed in this study. The performance of water–land discriminator based on NWSP is evaluated using water–land interface derived by the traditional waveform clustering. Water–land discriminator based on NWSP can reach an overall accuracy of 99.45%. The proposed method which needs IR and green laser point clouds is recommended for water–land discrimination with high-accuracy in coastal and inland waters.

Ocean-land waveform classification is a crucial step in processing of airborne LiDAR bathymetry (ALB) data and can be used for waterline extraction. Special laser waveforms induced by complex environments cause errors in labeling oceans and land based on differences in waveform features of infrared (IR) lasers. These special laser waveforms cannot be identified based on differences in waveform features or elevation and can only be corrected using spatial information. The traditional density clustering algorithm is too time-consuming and not easy to implement for the large amount of ALB data. In this paper, a correction method is proposed based on the continuity of laser pulses emitted by ALB. Experiments demonstrate that the proposed method corrects mislabeled waveforms with 57% reduction in the number of mislabeled waveforms compared to the K- means method and 99.3% reduction in time compared to the dual-clustering method.

Abstract Violence detection is a critical task aimed at identifying violent behavior in video by extracting frames and applying classification models. However, the complexity of video data and the suddenness of violent events present significant hurdles in accurately pinpointing instances of violence, making the extraction of frames that indicate violence a challenging endeavor. Furthermore, designing and applying high-performance models for violence detection remains an open problem. Traditional models embed extracted spatial features from sampled frames directly into a temporal sequence, which ignores the spatio-temporal characteristics of video and limits the ability to express continuous changes between adjacent frames. To address the existing challenges, this paper proposes a novel framework called ACTION-VST. First, a keyframe extraction algorithm is developed to select frames that are most likely to represent violent scenes in videos. To transform visual sequences into spatio-temporal feature maps, a multi-path excitation module is proposed to activate spatio-temporal, channel and motion features. Next, an advanced Video Swin Transformer-based network is employed for both global and local spatio-temporal modeling, which enables comprehensive feature extraction and representation of violence. The proposed method was validated on two large-scale datasets, RLVS and RWF-2000, achieving accuracies of over 98 and 93%, respectively, surpassing the state of the art.

Abstract Twitter bot detection is an important and meaningful task. Existing methods can be bypassed by the latest bots that disguise themselves as genuine users and evade detection by mimicking them. These methods also fail to leverage the clustering tendencies of users, which is the most important feature for detecting bots at the community level. Moreover, they neglect the implicit relations between users that contain crucial clues for detection. Furthermore, the user relation graphs, which are essential for graph-based methods, may be unreliable due to noise and incompleteness in datasets. To address these issues, a bot detection framework with graph structure learning is proposed. The framework constructs a heterogeneous graph with users and their relations, extracts multiple features to characterise user intent and establishes a feature similarity graph using metric learning. Implicit relations are discovered to derive an implicit relation graph. Additionally, a semantic relation graph is generated by aggregating relation semantics among users. The graphs are then fused and embedded into a Graph Transformer for training with partially known user labels. The framework demonstrated a 91.92% average detection accuracy on three real-world benchmark, outperforming state-of-the-art methods, while also showcasing the effectiveness and necessity of each module.

Abstract Pornographic video detection is of significant importance in curbing the proliferation of pornographic information on online video platforms. However, existing works often employ generic frame extraction methods that ignore the low-latency requirements of detection scenarios and the characteristics of pornographic videos. Additionally, existing detection methods have difficulties in detail characterization and semantic understanding, resulting in low accuracy. Therefore, this paper proposes an efficient pornographic video detection framework based on semantic and image enhancement. Firstly, a keyframe extraction method tailored for pornographic video detection is proposed to select representative frames. Secondly, a light enhancement method is introduced to facilitate accurate capture of pornographic visual cues. Moreover, a compression-reconstruction network is employed to eliminate adversarial perturbations, enabling models to obtain reliable features. Subsequently, YOLOv5 is introduced to locate and crop human targets in keyframes, reducing background interference and enhancing the expression of human semantic information. Finally, MobileNetV3 is employed to determine if the human targets contain pornographic content. The proposed framework is validated on the publicly available NPDI dataset, achieving an accuracy of 95.9%, surpassing existing baseline methods.

Long-term tracking is a commonly overlooked yet practical scenario in multi-object tracking. Handling occlusion and re-identifying long-lost targets are the main challenges for effective long-term tracking. In occlusion scenarios, both appearance and motion features can be unreliable, leading to association failure. For long-lost targets, predicting their long-term motion suffers from severe error accumulation, making the target re-identification challenging. In this paper, we propose a multi-object tracker called LTTrack for long-term tracking. For occlusion handling, we develop the Position-Based Association (PBA) module, which encodes relative and absolute positions as interaction and motion features for association. With interaction features, PBA can handle occlusion scenes where appearance and motion features are unreliable. For long-lost target re-identification, the Long-Term Motion (LTM) model is devised. By encoding long-term motion trends of targets for long-term motion prediction, LTM alleviates the error accumulation problem. Moreover, to prevent the erroneous deletion of long-lost tracks, we propose the Zombie Track Re-Match (ZTRM) strategy to re-identify long-lost targets so that they will neither be prematurely deleted nor disrupt the association of other tracks. Extensive experiments conducted on MOT17, MOT20, and DanceTrack demonstrate that LTTrack achieves performance comparable to state-of-the-art methods. The code and models are available at https://github.com/Lin-Jiaping/LTTrack.

Abstract A web shell is a backdoor used by hackers to control Web servers and perform privilege escalation, and thus it is crucial to detect web shells effectively. However, the detection of obfuscated web shells has always been a challenge. Inspired by adversarial training methods in the field of computer vision, this paper proposes a generative adversarial network (GAN)-based web shell detection model training framework. Since there has been no method that can generate obfuscated web shells effectively, a generator based on the genetic algorithm, which combines and optimizes the pre-set obfuscation methods, is used to obtain new obfuscation combinations and generate obfuscated samples. The whole proposed framework is named the CWSOGG. When training the detection model, the generator generates web shells that can bypass the discriminator, and the discriminator catches the features of obfuscated samples. Through the adversarial training of the discriminator and generator, the detection model improves its ability to detect obfuscated web shells. To verify the proposed framework is flexible to different models, the discriminator based on four main neural networks has been implemented. Meanwhile, to build complete feature extraction models, both statistical and semantic features are extracted. Due to the lack of web shell data, a clean dataset containing 4,375 web shells is constructed and used to evaluate the CWSOGG. The results have shown that the detection accuracy of each model increases by 86.71% on the generated obfuscated web shells on average and by 7.50% on the simulated real-world obfuscated web shells on average.

In this paper, we introduce VCSL (Video Copy Segment Localization), a new comprehensive segment-level annotated video copy dataset. Compared with existing copy detection datasets restricted by either video-level annotation or small-scale, VCSL not only has two orders of magnitude more segment-level labelled data, with 160k realistic video copy pairs containing more than 280k localized copied segment pairs, but also covers a variety of video categories and a wide range of video duration. All the copied segments inside each collected video pair are manually extracted and accompanied by precisely annotated starting and ending timestamps. Alongside the dataset, we also propose a novel evaluation protocol that better measures the prediction accuracy of copy overlapping segments between a video pair and shows improved adaptability in different scenarios. By benchmarking several baseline and state-of-the-art segment-level video copy detection methods with the proposed dataset and evaluation metric, we provide a comprehensive analysis that uncovers the strengths and weaknesses of current approaches, hoping to open up promising directions for future works. The VCSL dataset, metric and benchmark codes are all publicly available at https://github.com/alipay/vCSL.

The grim reality of fatigued driving-caused traffic accidents indicates the urgent need to establish a robust fatigued driving detection method to prevent drivers from falling asleep while driving. Previous studies on this topic have little consideration on the sequence-level characteristics of driver’s behaviors, furthermore, many of them only focus on one type of feature, which limits the accuracy and flexibility. Besides, most existing computer vision-based studies perform experiments on high-performance hardware like GPUs, which can not demonstrate their performance in the reality, as it is impractical to equip such high-end computing hardware in vehicles. This paper proposes a novel low-power hardware-friendly fatigued driving detection method based on multi-feature fusion and time series classification. The system utilizes a camera installed on the vehicle for driver monitoring. Continuous frames of a specific duration are a detection sequence. Multiple physical features including head orientation, face landmark coordinates, EFV (Eye Feature Vector), and MFV (Mouth Feature Vector) are extracted from each frame. A neural network optimized for low-power hardware is used to classify the multivariate time series samples denoting the driver’s state into two classes: fatigued or alert. Sequence-level analysis reduces the false positive rate by avoiding being affected by accidental symptoms. Multi-feature enhances the capability to give the right detection results in complex and changeable driving scenarios. The proposed method is comprehensively evaluated on an ARM platform-based device simulating the low-power in-vehicle hardware environment. Experiments show our models surpass existing studies by up to 11.49% on accuracy and achieve up to 1.41× speedup, proving the significant advantages of the proposed method.

In recent years, microblog systems such as Twitter and Sina Weibo have averaged multimillion active users. On the other hand, the microblog system has become a new means of rumor-spreading platform. In this paper, we investigate the machine-learning-based rumor identification approaches. We observed that feature design and selection has a stronger impact on the rumor identification accuracy than the selection of machine-learning algorithms. Meanwhile, the rumor publishers' behavior may diverge from normal users', and a rumor post may have different responses from a normal post. However, mass behavior on rumor posts has not been explored adequately. Hence, we investigate rumor identification schemes by applying five new features based on users' behaviors, and combine the new features with the existing well-proved effective user behavior-based features, such as followers' comments and reposting, to predict whether a microblog post is a rumor. Experiment results on real-world data from Sina Weibo demonstrate the efficacy and efficiency of our proposed method and features. From the experiments, we conclude that the rumor detection based on mass behaviors is more effective than the detection based on microblogs' inherent features.

We develop an expectation-maximization algorithm with local adaptivity for image segmentation and classification. The key idea of our approach is to combine global statistics extracted from the Gaussian mixture model or other proper statistical models with local statistics and geometrical information, such as local probability distribution, orientation, and anisotropy. The combined information is used to design an adaptive local classification strategy that improves the robustness of the algorithm and also keeps fine features in the image. The proposed methodology is flexible and can be easily generalized to deal with other inferred information/quantities and statistical methods/models.

The goal of this paper is to develop a framework for designing good data repositories for Web applications. The central theme of our approach is to employ statistical methods to predict quality metrics. These prediction quantities can be used to answer important questions such as: How soon should the local repository be synchronized to have a quality of at least 90% precision with certain confidence level? Suppose the local repository was synchronized three days ago, how many objects could have been deleted at the remote source since then?

In this paper, a novel approach is proposed for estimating traffic matrices. Our method, called PamTram for PArtial Measurement of TRAffic Matrices, couples lightweight origin-destination (OD) flow measurements along with a computationally lightweight algorithm for producing OD estimates. The first key aspect of our method is to actively select a small number of informative OD flows to measure in each estimation interval. To avoid the heavy computation of optimal selection, we use intuition from game theory to develop randomized selection rules, with the goals of reducing errors and adapting to traffic changes. We show that it is sufficient to measure only one flow per measurement period to drastically reduce errors-thus rendering our method lightweight in terms of measurement overhead. The second key aspect is an explanation and proof that an Iterative Proportional Fitting algorithm approximates traffic matrix estimates when the goal is a minimum mean-squared error; this makes our method lightweight in terms of computation overhead. A one-step error bound is provided for PamTram that bounds the average error for the worst scenario. We validate our method using data from Sprint's European Tier-1 IP backbone network and demonstrate its consistent improvement over previous methods.

The convergence rates of generalized iterative scaling (GIS) and improved iterative scaling (IIS) algorithms for fitting maximum entropy (ME) models are investigated and also a particular linear dynamic system monitoring with partial active measurements is studied. An information-theoretic based measurement scheme is derived to select informative hidden states, which is validated on a problem of origin-destination matrix estimation for Internet traffic.

Network monitoring and diagnosis are key to improving network performance. The difficulties of performance monitoring lie in today's fast growing Internet, accompanied by increasingly heterogeneous and unregulated structures. Moreover, these tasks become even harder since one cannot rely on the collaboration of individual routers and servers to directly measure network traffic. Even though the aggregatory nature of possible network measurements gives rise to inverse problems, existing methods for solving inverse problems are usually computationally intractable or statistically inefficient. In this paper, a pseudo likelihood approach is proposed to solve a group of network tomography problems. The basic idea of pseudo likelihood is to form simple subproblems and construct a product of marginal likelihood of subproblems by the ignoring their dependences. As a result, it keeps a good balance between the computational complexity and the statistical efficiency of the parameter estimation. Some statistical properties of the pseudo likelihood estimator, such as consistency and asymptotic normality, are established. A pseudo expectation-maximization (EM) algorithm is developed to maximize the pseudo log-likelihood function. Two examples with simulated or real data are used to illustrate the pseudo likelihood proposal: (1) internal link delay distribution inference through multicast end-to-end measurements; (2) origin-destination matrix estimation through link traffic counts.

This paper describes Iterated Systems' submission to the MPEG-4 committee in January 1996. A system for compressing video is presented which expresses predictive frames of a sequence by means of motion vectors applied to variable-size blocks with a constant intensity adjustment. The motion vectors are organized into a partial quadtree, which allows incomplete splitting of blocks into zero to four quadrants. The motion vectors, mean intensity offsets, and partial quadtree structure are arrived at by a joint optimization process which may be carried out in a bottom-up fashion. A form of generalized overlapped block motion compensation applicable at all block sizes is presented, and a method for encoding segmented video is shown. The performance of the algorithm is compared with that of Telenor's H.263 for three of the MPEG-4 test sequences which show that the former is comparable or superior to the latter for low bit rates.

Much research has focused on the problem of estimating the model order of autoregressive moving average (ARMA) processes. The most well-known of the proposed solutions for this problem include the final prediction error (FPE), Akaike information criterion (AIC), and minimum description length (MDL). A new approach for model order determination based on the MDL criterion is proposed and shown to depend on the minimum eigenvalue of a covariance matrix derived from the observed data. As a result, a new selection procedure for estimating the model order via MDL is proposed. Examples that illustrate the significantly improved accuracy of the proposed technique are given.>