检索结果-内蒙古大学图书馆

Deepfake Detection System Using a Hybrid Model

学校读者我要写书评

暂无评论

Deepfake Detection System Using a Hybrid Model

International Conference on computing and Networking technology (ICCNT)

作者： Sudharson S Priyanka Kokil Sasikala D Penta Aravinda Swamy Computer Science and Engineering Vellore Institute of Technology Chennai India Electronics and Communication Engineering Indian Institute of Information Technology Design and Manufacturing Chennai India Amrita School of Computing Computer Science and Engineering (AIE) Chennai India

There has been a rapid advancement in deep fake technology in recent years, which makes it increasingly difficult to detect fake images, especially those involving human faces, which have become difficult to detect. The traditional methods of detecting deep fake images are usually based on a human-crafted feature or a rule-based system, which can be easily beaten with simple cheating techniques. Intending to detect deepfake face images, we present a novel method that uses CapsNet and EfficientNet as feature extractors to detect deepfake face images. They are well suited to detect deepfake images that have undergone geometric transformations or variations as they can learn the spatial relationship between those features. Using EfficientNet, a powerful and efficient deep learning architecture, as a feature extractor, we extract high-level characteristics from an input image to create a more accurate image *** on our experiments,it was found that this method demonstrated the capability to detect various types of deep fake images with an accuracy rate of 99.64%, including those fabricated through face swapping and re-enacting. This method could be applied to a wide range of real-world applications, including in media forensics, law enforcement, and online media verification, so that it could be applied in many different contexts.

关键词：

Class Imbalance Pill Image Classification via Ensemble Learning

学校读者我要写书评

暂无评论

Class Imbalance Pill Image Classification via Ensemble Learn...

IEEE Conference on Systems, Process & Control (ICSPC)

作者： Hui-Zong Chee Jit-Yan Lim Yong-Xuan Tan Kian-Ming Lim Chin-Poo Lee Pa-Pa-Min Faculty of Information Science & Technology (FIST) Multimedia University Melaka Malaysia School of Computer Science University of Nottingham Ningbo China Ningbo China School of AI and Advanced Computing Xi’an Jiaotong-Liverpool University Jiangsu Province China

ISBN: (数字)9798350391398

ISBN: (纸本)9798350391404

Accurate identification of pill shapes is crucial for medication management and safety, acting as a safeguard against medication errors, especially with the increasing variety of pills that pose risks to patients. To address this, we propose a novel model for class imbalance pill image classification with the use of transfer learning and ensemble learning. Our proposed ensemble model integrates custom built Convolutional Neural Networks (CNN) and pre-trained EfficientNet-B2, both of which excel in image classification tasks. The CNN architecture employs multiple convolutional and pooling layers to identify pill characteristics, while EfficientNet enhances performance through its pre-trained capabilities. The proposed ensemble model demonstrated exceptional performance, achieving accuracies of 99.40% on the NIH Pill Image Dataset and 100% on the PillBR Dataset, respectively.

关键词： Accuracy Shape Transfer learning Safety Convolutional neural networks Ensemble learning Image classification Biomedical imaging

Shield-U: Safeguarding Traffic Sign Recognition Against Perturbation Attacks

学校读者我要写书评

暂无评论

Shield-U: Safeguarding Traffic Sign Recognition Against Pert...

IEEE International Conference on Trust, Security and Privacy in computing and Communications (TrustCom)

作者： Shengmin Xu Jianfei Sun Hangcheng Cao Yulan Gao Ziyang He Cong Wu College of Computer and Cyber Security Fujian Normal University China School of Computing and Information Systems Singapore Management University Singapore Department of Computer Science City University of Hong Kong China Division of Information Science and Engineering KTH Royal Institute of Technology Sweden School of Cyber Science and Engineering Zhengzhou University China School of Computer Science and Engineering Nanyang Technological University Singapore

ISBN: (数字)9798331506209

ISBN: (纸本)9798331506216

Traffic sign recognition systems are crucial for the navigation and situation awareness of autonomous vehicles. They leverage deep learning technologies to swiftly and accurately identify traffic signs, even in the most challenging traffic environments. However, security researchers have uncovered a critical vulnerability in these systems: learning-based TSRs are particularly susceptible to physical-world perturbation attacks. Through subtle modifications (i.e., attaching well-designed patches on traffic signs), attackers can deceive the recognition system into making erroneous judgments, which can further lead to serious traffic accidents. Although several defense mechanisms have been proposed to enhance the security of sign recognition systems, these solutions generally target only specific types of malicious perturbations and thus lack robustness. To address this issue, we present a robust defense mechanism named Shield-U, which restores traffic sign images contaminated by physical patch perturbations, providing credible data for the recognition model. In the process of implementing Shield-U, we first design a feature difference-aware perturbation generator that outputs potential sign contamination patterns. Incorporating generated perturbations during the training phase enables our restoration model to gain sufficient understanding of diverse perturbation types, thus enhancing its ability to repair various perturbed signs. Following this, we build an attention-driven restoration network to repair sign images. Finally, we evaluate the effectiveness of Shield-U using widely used sign recognition models and public datasets. The results demonstrate that our defense mechanism excels in resisting potential perturbations, increasing the average sign recognition accuracy by 50.4%.

关键词： Training Image recognition Target recognition Perturbation methods Maintenance engineering Robustness Image restoration Safety Security Public transportation

Deep CNN based droplet deposition segmentation for spray distribution assessment 27

学校读者我要写书评

暂无评论

Deep CNN based droplet deposition segmentation for spray dis...

27th International Conference on Automation and computing, ICAC 2022

作者： Chen, Tao Meng, Yanhua Su, Jinya Liu, Cunjia University of Essex School of Computer Science and Electronic Engineering Colchester United Kingdom Anyang Institute of Technology School of Mechanical Engineering Anyang China University of Aberdeen Department of Computing Science Aberdeen United Kingdom Loughborough University Department of Aeronautical and Automotive Engineering Loughborough United Kingdom

ISBN: (数字)9781665498074

ISBN: (纸本)9781665498074

Pesticides have been widely used in the cultivation of crops to enhance their production, however, incorrect application of pesticides will result in yield loss, product waste, environmental pollution among many others. Therefore, timely evaluating spray distribution of intelligent sprayers plays a pivotal role in the appropriate delivery of pesticides to the crop. The exiting approaches based on water-sensitive paper (WSP) either involve a relatively tedious and labor-intensive procedure, or have a high requirement on WSP image taking. So in this study we aim to conduct spray distribution assessment in the field based on mobile devices. To this end, the key issue of droplet deposition segmentation under natural imaging environments is addressed. WSPs with food dye droplets are first collected in the field by mobile phones. Then an image dataset on droplet deposition segmentation is created via thresholding approach with human supervision. Then four popular deep convolutional neural network (CNN) based segmentation algorithms are applied for droplet deposition segmentation so that spray distribution can be assessed. Comparative experiments show that UNeXt network is the best one in consideration of accuracy, inference time and network size. © 2022 IEEE.

关键词： Semantic Segmentation

PGD-Imp: Rethinking and Unleashing Potential of Classic PGD with Dual Strategies for Imperceptible Adversarial Attacks

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Li, Jin Yu, Zitong He, Ziqiang Wang, Z. Jane Kang, Xiangui Guangdong Key Lab of Information Security School of Computer Science and Engineering Sun Yat-Sen University China School of Computing and Information Technology Great Bay University China Electrical and Computer Engineering Dept University of British Columbia Canada

Imperceptible adversarial attacks have recently attracted increasing research interests. Existing methods typically incorporate external modules or loss terms other than a simple lp-norm into the attack process to achieve imperceptibility, while we argue that such additional designs may not be necessary. In this paper, we rethink the essence of imperceptible attacks and propose two simple yet effective strategies to unleash the potential of PGD, the common and classical attack, for imperceptibility from an optimization perspective. Specifically, the Dynamic Step Size is introduced to find the optimal solution with minimal attack cost towards the decision boundary of the attacked model, and the Adaptive Early Stop strategy is adopted to reduce the redundant strength of adversarial perturbations to the minimum level. The proposed PGD-Imperceptible (PGD-Imp) attack achieves state-of-the-art results in imperceptible adversarial attacks for both untargeted and targeted scenarios. When performing untargeted attacks against ResNet-50, PGD-Imp attains 100% (+0.3%) ASR, 0.89 (-1.76) l2 distance, and 52.93 (+9.2) PSNR with 57s (-371s) running time, significantly outperforming existing methods. Copyright © 2024, The Authors. All rights reserved.

关键词： Adversarial machine learning

Learning Adversarial Semantic Embeddings for Zero-Shot Recognition in Open Worlds

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Li, Tianqi Pang, Guansong Bai, Xiao Zheng, Jin Zhou, Lei Ning, Xin School of Computer Science and Engineering Beihang University China School of Computing and Information Systems Singapore Management University Singapore School of Computer Science and Artificial Intelligence Wuhan University of Technology China Institute of Semiconductors Chinese Academy of Sciences China

Zero-Shot Learning (ZSL) focuses on classifying samples of unseen classes with only their side semantic information presented during training. It cannot handle real-life, open-world scenarios where there are test samples of unknown classes for which neither samples (e.g., images) nor their side semantic information is known during training. Open-Set Recognition (OSR) is dedicated to addressing the unknown class issue, but existing OSR methods are not designed to model the semantic information of the unseen classes. To tackle this combined ZSL and OSR problem, we consider the case of "Zero-Shot Open-Set Recognition" (ZS-OSR), where a model is trained under the ZSL setting but it is required to accurately classify samples from the unseen classes while being able to reject samples from the unknown classes during inference. We perform large experiments on combining existing state-of-the-art ZSL and OSR models for the ZS-OSR task on four widely used datasets adapted from the ZSL task, and reveal that ZS-OSR is a non-trivial task as the simply combined solutions perform badly in distinguishing the unseen-class and unknown-class samples. We further introduce a novel approach specifically designed for ZS-OSR, in which our model learns to generate adversarial semantic embeddings of the unknown classes to train an unknowns-informed ZS-OSR classifier. Extensive empirical results show that our method 1) substantially outperforms the combined solutions in detecting the unknown classes while retaining the classification accuracy on the unseen classes and 2) achieves similar superiority under generalized ZS-OSR settings. Our code is available at https://***/lhrst/ASE. © 2023, CC BY.

关键词： Zero-shot learning

Advancements in cheating detection algorithms in FPS games

学校读者我要写书评

暂无评论

Advancements in cheating detection algorithms in FPS games

2024 International Conference on computer Application and Information Security, ICCAIS 2024

作者： Xu, Ziwen Yu, Ruotong Liang, Yan Chen, Zihan Wan, Ye School of Computer Science and Technology Tongji University Shanghai201804 China Faculty of Science and Technology Beijing Normal University-Hong Kong Baptist University United International College Guangdong Zhuhai519087 China Luddy School of Informatics Computing and Engineering Indiana University of Bloomington Indiana BloomingtonIN47405 United States Guangdong Guangya High School Guangdong Guangzhou510000 China School of Computer Science and Technology Northeastern University Hebei Qinhuangdao066000 China

ISBN: (纸本)9781510689312

Over the past two decades, the increasing variety of cheating techniques in FPS games has led to the development of cheating detection algorithms. This literature review provides a chronological review of advancements in cheating detection techniques. Early methods, such as avatar trajectory-based detection, focused on behavioral analysis and identified bots by analyzing movement patterns. A remote anomaly-based system called Fides was presented to detect client-side manipulations with client-server architectures. With the emergence of machine learning, a variety of supervised learning methods have been introduced into the field of game detection to achieve real-time detection. More recent studies have been more innovative. Traps like the "bait-target" technique were introduced to detect cheating mechanisms that mimic human behavior. Methods combining deep learning and computer vision were proposed to detect complex cheating based on visual images. Overall, this review shows the diversity of cheat detection algorithms that have evolved to maintain the fairness of competitive gaming environments. © The Authors.

关键词： Supervised learning

HUWSOD: Holistic Self-training for Unified Weakly Supervised Object Detection

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Cao, Liujuan Lin, Jianghang Hong, Zebo Shen, Yunhang Lin, Shaohui Chen, Chao Ji, Rongrong Media Analytics and Computing Lab School of Informatics Xiamen University 361102 China Youtu Lab Tencent200233 China School of Computer Science and Technology East China Normal University 200062 China

As an emerging problem in computer vision, weakly supervised object detection (WSOD) aims to use only image-level annotations to train object detectors. Most WSOD methods rely on traditional object proposals to generate candidate regions and are confronted with unstable training, which easily gets stuck in a poor local optimum. In this paper, we propose a unified and high-capacity WSOD network with holistic self-training framework, termed HUWSOD, which is self-contained, and requires no external modules or additional supervision. To this end, the proposed framework innovates in two perspectives, i.e., unified network structure and self-training scheme, both of which are rarely touched in WSOD before. First, we design a self-supervised proposal generator and an autoencoder proposal generator with multi-rate resampling pyramid to hypothesize object locations, which replace the traditional object proposals and enable the strictly end-to-end training and inference for WSOD. Second, we introduce holistic self-training scheme that consists of step-wise entropy minimization and consistency-constraint regularization, which refine both detection scores and coordinates progressively, as well as enforcing consistency to match predictions produced from stochastic augmentations of the same image. Extensive experiments on PASCAL VOC and MS COCO show that the proposed HUWSOD performs competitively to the state-of-the-art WSOD methods, while getting rid of offline proposals and additional data. Moreover, the upper-bound performance of HUWSOD with class-agnostic ground-truth bounding boxes approaches Faster R-CNN, which demonstrates that HUWSOD is capable to achieve fully-supervised accuracy. Our work also brings one important finding that random initialized boxes, although drastically different from the offline well-designed object proposals, are also effective object candidates for WSOD training. The code is available at: https://***/shenyunhang/HUWSOD. Copyright © 2024, Th

关键词： Object detection

G2Face: High-Fidelity Reversible Face Anonymization via Generative and Geometric Priors

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Yang, Haoxin Xu, Xuemiao Xu, Cheng Zhang, Huaidong Qin, Jing Wang, Yi Heng, Pheng-Ann He, Shengfeng The School of Computer Science and Engineering South China University of Technology Guangzhou China Guangdong Engineering Center for Large Model and GenAI Technology State Key Laboratory of Subtropical Building and Urban Science China The Centre for Smart Health The Hong Kong Polytechnic University Hong Kong The School of Future Technology South China University of Technology Guangzhou China The School of Computer Science and Network Security Dongguan University of Technology Dongguan China The Department of Computer Science and Engineering The Chinese University of Hong Kong Hong Kong The School of Computing and Information Systems Singapore Management University Singapore

Reversible face anonymization, unlike traditional face pixelization, seeks to replace sensitive identity information in facial images with synthesized alternatives, preserving privacy without sacrificing image clarity. Traditional methods, such as encoder-decoder networks, often result in significant loss of facial details due to their limited learning capacity. Additionally, relying on latent manipulation in pre-trained GANs can lead to changes in ID-irrelevant attributes, adversely affecting data utility due to GAN inversion inaccuracies. This paper introduces G2Face, which leverages both generative and geometric priors to enhance identity manipulation, achieving high-quality reversible face anonymization without compromising data utility. We utilize a 3D face model to extract geometric information from the input face, integrating it with a pre-trained GAN-based decoder. This synergy of generative and geometric priors allows the decoder to produce realistic anonymized faces with consistent geometry. Moreover, multi-scale facial features are extracted from the original face and combined with the decoder using our novel identity-aware feature fusion blocks (IFF). This integration enables precise blending of the generated facial patterns with the original ID-irrelevant features, resulting in accurate identity manipulation. Extensive experiments demonstrate that our method outperforms existing state-of-the-art techniques in face anonymization and recovery, while preserving high data utility. Code is available at https://***/Harxis/G2Face. Copyright © 2024, The Authors. All rights reserved.

关键词： Anonymity