检索结果-内蒙古大学图书馆

Automatic Face Expression Neutralisation and Posture Correction via 3D Contrastive Learning

Procedia Computer Science 2023年 225卷 269-278页

作者： Grina Wiem Douik Ali University of Sousse Networked Objects Control and Communication Systems Laboratory National School of Engineers of Sousse Sousse Technology Center Sahloul Belt Road 4054 University of Monastir National Engineering School of Monastir Rue Ibn Jazzar Monastir 5035 Tunisia University of Sousse Networked Objects Control and Communication Systems Laboratory National School of Engineers of Sousse Sousse Technology Center Sahloul Belt Road 4054 Tunisia

The performance of recognition systems can be significantly affected by various factors, with facial expression poses and lighting changes being the main confounding factors. In order to minimize their impact, we propose an intriguing method in this study that enables the generation of high-quality images specifically tailored to the target domain. Our objective is to utilize disentangled representation to effectively model the decomposition of data variations and generate neutral facial expression images with frontal posture and adaptive illumination. To achieve this, we incorporate 3D priors in the adversarial learning during the training process, simulating the generation of an analytical 3D face deformation as well as rendering operations. Additionally, we employ contrastive learning to control the disentanglement of the generated faces while preserving the essential properties of facial features. This technique enables us to learn an embedding space in which similar data samples are represented closely, while distinct samples are kept far apart from each other. Furthermore, we conduct an analysis of the learned latent space and introduce several other significant properties that enhance the reinforcement of factor disentanglement. These properties include an imitation learning algorithm, which facilitates the acquisition of meaningful patterns and characteristics.

关键词： generative adversarial network expression neutralization disentanglement 3D contrastive learning

来源：评论

学校读者我要写书评

暂无评论

Sliding mode control of quadrotor based on differential flatness

Sliding mode control of quadrotor based on differential flat...

引用

2018 International Conference on control, Automation and Diagnosis, ICCAD 2018

作者： Abadi, Amine Hadj Brahim, Anis Ben Mekki, Hassen Amraoui, Adnen El Ramdani, Nacim Networked Objects Control and Communication Systems Laboratory National Engineering School of Sousse University of Sousse Tunisia University of Sousse National Engineering School of Sousse Tunisia Univ. Orléans INSA CVL PRISME EA 4229 OrléansF45072 France

ISBN: (纸本)9781538654071

In this work, the differential flatness property and the sliding mode controller design are proposed for a quadrotor in order to track a predefined reference trajectory despite the presence of uncertain parameters. Firstly, differential flatness is utilized to resolve the problem for trajectory generation and tracking for the quadrotor. Next, a sliding controller is combined with a flatness design to guarantee the robustness of the tracking strategy. The numerical simulations of quadrotor system are done in order to evaluate the performance of the suggested control scheme. © 2018 IEEE.

关键词： Sliding mode control

来源：评论

学校读者我要写书评

暂无评论

CNL-ResUnet: A novel deep learning architecture for stroke image segmentation

CNL-ResUnet: A novel deep learning architecture for stroke i...

引用

International Conference on Advanced Technologies for Signal and Image Processing (ATSIP)

作者： Nesrine Jazzar Besma Mabrouk Ali Douik Networked Objects Control and Communication Systems National Engineering School of Sousse University of Sousse Tunisia National Engineering School of Sfax University of Sfax Tunisia Department of Electrical and Computer Engineering Advanced Technologies for Medicine and Signals ATMS National Engineers School Sfax University Tunisia Networked Objects Control and Communication Systems National Engineering School Sousse Tunisia

ISBN: (数字)9798350351484

ISBN: (纸本)9798350351491

Segmentation of stroke lesions holds significant importance in enabling timely diagnosis and subsequent treatment by medical professionals. Automating this process presents technical challenges given the diverse appearance and evolving nature of lesions, discrepancies among medical experts, limited dataset availability, and the necessity for multiple MRI modalities for comprehensive imaging. This research presents an innovative method called CNL-ResUNet, specifically designed for 2D stroke image segmentation. CNL-ResUNet ingeniously incorporates residual blocks into its architecture, capitalizing on their innate capacity to facilitate seamless information flow across layers, thereby empowering the network to efficiently distill complex patterns from sparse datasets. It integrates customized skip connections to reduce semantic disparities between the encoder and decoder layers. Additionally, the architecture incorporates an innovative Classifier and Localizer (CNL) module, providing supplementary classification and localization insights with enhanced accuracy. By combining these insights with the segmentation results, CNL-ResUNet adeptly reduces the occurrence of false positives and false negatives. To evaluate the model’s performance, experiments were conducted utilizing the SISS dataset from the Ischemic Stroke Lesion Segmentation Challenge 2015. The outcomes from these experiments highlight the superiority of our approach, revealing significant qualitative and quantitative improvements across standard evaluation metrics like the Dice coefficient and Accuracy. Our proposed network demonstrates potential for advancing automated tools in stroke segmentation.

关键词： Deep learning Training Image segmentation Accuracy Semantics Stroke (medical condition) Robustness

来源：评论

学校读者我要写书评

暂无评论

Comparison of RNN-Based Models for Sign Language Recognition and Translation

Comparison of RNN-Based Models for Sign Language Recognition...

引用

International Conference on control, Decision and Information Technologies (CoDIT)

作者： Marah El Askri Riadh Bchir Hend Basly Fatma Ezahra Sayadi National Engineering School of Sousse (ENISO) Laboratory of Networked Objects Control and Communication Systems (NOCCS) University of Sousse Sousse Tunisia Faculty of Sciences of Monastir University of Monastir Monastir Tunisia

ISBN: (数字)9798350373974

ISBN: (纸本)9798350373981

The recognition of human activities using vision based methods has become a key area of interest in video analytics research. In the past years, several sophisticated deep learning (DL) algorithms have surfaced, designed to recognize complex human actions depicted in video streams. These approaches have exhibited remarkable competence in tasks related to video analytics. This paper explores the field in greater depth by centering its focus on Sign Language (SL) as a unique and complex form of human activity. SL plays is pivotal in facilitating communication for people who are deaf or hard of hearing, enabling them to interact and engage. The main idea is to recognize Arabic Sign Language especially Tunisian accent, and subsequently translate it into spoken language utilizing artificial intelligence (AI) methods. The goal of this work is to facilitate communication between individuals who are deaf and those who are not. So, this paper presents the comparative results of four deep-learning models: Long Short-Term Memory (LSTM), Bidirectional Long Short-Term Memory (Bi-LSTM), Gated Reccurent Unit (GRU), and Bidirectional Gated Reccurent Unit (Bi-GRU), aiming to improve the recognition accuracy of dynamic signs. The results indicate that the BiLSTM outperforms other variants, showcasing an accuracy of 98%.

关键词： Sign language Accuracy Recurrent neural networks Visual analytics Logic gates Streaming media Artificial intelligence Information technology Long short term memory Streams

来源：评论

学校读者我要写书评

暂无评论

Sliding Mode control of Quadrotor based on Differential Flatness

Sliding Mode Control of Quadrotor based on Differential Flat...

引用

International Conference on control, Automation and Diagnosis (ICCAD)

作者： Amine Abadi Anis Ben Hadj Brahim Hassen Mekki Adnen El Amraoui Nacim Ramdani National Engineering School Of Sousse Networked Objects Control and Communication Systems Laboratory Tunisia National Engineering School Of Sousse University of Sousse Univ. Orléans Orléans France

ISBN: (纸本)9781538654088

关键词： Trajectory Mathematical model Sliding mode control Robustness Drones Atmospheric modeling

来源：评论

学校读者我要写书评

暂无评论

Deep Learning For Intra Frame Coding

Deep Learning For Intra Frame Coding

引用

International Conference on Engineering and Emerging Technologies (ICEET)

作者： Maraoui Amna Werda Imen Sayadi Fatma Ezahra Electronics and Microelectronics Laboratory University of Monastir Monastir Tunisia Electronic and Information Technology Laboratory University of Sfax Sfax Tunisia Networked objects control and communication systems Laboratory University of Sousse Sousse Tunisia

ISBN: (纸本)9781665427159

The evolution of the coding unit module from the High Efficiency Video Coding (HEVC) video standards to the Joint Exploration Model (JEM) extensively enhanced compression performance while severely increases coding complexity caused by the brute force search built on Rate Distortion Optimization (RDO). Effectively, compared to the predecessor HEVC standard that makes use of the quad-tree (QT) block partitioning module, the novel quad-tree binary-tree (QTBT) block partitioning process proposed within the JEM encoder uses additional block sizes and shapes which induces additional flexibility. In this paper, we suggest a deep convolutional neural network (CNN) based method to reduce the block partitioning module complexity for both HEVC and JEM encoders at all intra-configurations. The results show that the CNN approach leads to better optimization performance with the HEVC encoder reaching up to 59%. However, the CNN model is more robust with several JEM versions.

关键词： Deep learning Shape Rate-distortion Encoding Complexity theory Convolutional neural networks Optimization

来源：评论

学校读者我要写书评

暂无评论

A Comparative Study of PCA and LDA for Dimensionality Reduction in a 4-Way Classification Framework

Research Square

引用

Research Square 2024年

作者： Mabrouk, Besma Jazzar, Nessrine Sallemi, Lamia Hamida, Ahmed Ben Advanced Technologies for Medicine and Signals ATMS Department of Electrical and Computer Engineering National Engineers School Sfax University Tunisia Networked Objects Control and Communication Systems National Engineering School Sfax Tunisia

Alzheimer’s disease (AD), recognized as the second-most impactful neurological disorder and currently incurable, stands as the leading cause of dementia. An imperative research focus is efficiently diagnosing the stages of patients, distinguishing early or late Mild Cognitive Impairment and AD from those with normal cognitive function. Advancements in anatomical and diffusion-weighted imaging, coupled with machine learning techniques, have significantly progressed in this predictive domain. However, in real-world trials, datasets often contain numerous features, and the curse of dimensionality can introduce challenges such as increased computational complexity, overfitting, and diminished model interpretability. To address these issues, the present study explores the efficacy of Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) as dimensionality reduction techniques. LDA, a supervised technique emphasizing class separability, surpasses PCA, particularly in selecting features that significantly contribute to discriminating between classes. The 3D-LDA features obtained were subsequently assessed across various machine learning algorithms, leading to the establishment of a 4-way classification framework that utilized the K-Nearest Neighbors model. The outcome of this evaluation yielded an impressive accuracy rate of 87% in predicting the four different classes. © 2024, CC BY.

关键词： Nearest neighbor search

来源：评论

学校读者我要写书评

暂无评论

Early Diagnosis of Diabetic Retinopathy using Random Forest Algorithm

Early Diagnosis of Diabetic Retinopathy using Random Forest ...

引用

International Conference on Advanced Technologies for Signal and Image Processing (ATSIP)

作者： Nihel ZAABOUB Ali DOUIK Networked Objects Control and Communication Systems (NOCCS) Laboratory ENISo University of Sousse National Engeneering School of Tunis (ENIT) University of Tunis El Manar National Engeneering School of Sousse (ENISo) University of Sousse

ISBN: (数字)9781728175133

ISBN: (纸本)9781728175140

The diabetic retinopathy is one of the most frequent causes of visual damage and vision loss. It can cause blindness in the absence of the diagnosis and the treatment. The automatic detection of the hard exudate in color fundus retinal images is an important task to early diagnosis the diabetic retinopathy. In this paper, a hard exudate detection algorithm is proposed. It is based on the application of a learning method to retinal image with removed optic disk. This paper proposes the use of Random Forest algorithm with a specific parameter from which a binary mask of exudate is obtained after intensity thresholding. It achieves 91.40% for sensitivity and 94.38% for the accuracy.

关键词： Diabetes Retinopathy Retina Sensitivity Optical sensors Optical imaging Vegetation

来源：评论

学校读者我要写书评

暂无评论

FFGAN: An Auto-supervised Approach for Frontal Face Generation Via Disentangled Contrastive Learning

引用

Procedia Computer Science 2024年 246卷 3644-3653页

作者： Grina Wiem Douik Ali National Engineering School of Monastir University of Monastir Rue Ibn Jazzar 5035 Monastir Tunisia Networked Objects Control and Communication Systems Laboratory Sousse Technology Centre National School of Engineers of Sousse University of Sousse Sahloul Belt Road 4054 Sousse Tunisia

This work proposes a novel generative model, FF-GAN (Frontal Face Generative Adversarial Network), for generating high-quality and diverse frontal faces. FF-GAN utilizes contrastive learning to effectively learn the underlying representation of frontal faces from unlabeled image datasets. This approach allows the model to capture the essential characteristics of frontal faces without the need for explicit pose annotations. We evaluate the performance of FF-GAN using established metrics like FID (Fréchet Inception Distance), IS (Inception Score), and SSIM (Structural Similarity Index Measure). The results demonstrate that FF-GAN achieves superior performance compared to existing methods, generating highly realistic and visually appealing frontal faces with exceptional structural coherence. This research contributes to the field of facial image generation by introducing an effective unsupervised learning approach based on contrastive learning for generating high-quality frontal faces.

关键词： Unsupervised Facial Pose Analysis Contrastive Learning Disentanglement Pose Estimation

来源：评论

学校读者我要写书评

暂无评论

Hybrid approach for speaker recognition based on formant and pitch extraction 22

Hybrid approach for speaker recognition based on formant and...

引用

22nd International Conference on Cyberworlds, CW 2023

作者： Boujnah, Sana Ferjaoui, Radhia Khalifa, Anouar Ben University of Tunis El Manar Tunisia National Engineering School of Tunis Tunisia NOCCS-Laboratory of Networked Objects Control and Communication Systems Eniso Tunisia Istmt Tunis Tunisia Université de Sousse Ecole Nationale d'Ingénieurs de Sousse LATIS-Laboratory of Advanced Technology and Intelligent Systems Sousse4023 Tunisia Université de Jendouba Institut National des Technologies et des Sciences du Kef Le Kef7100 Tunisia

ISBN: (纸本)9798350315653

Human voice is an ideal data source for identifying people in many applications. Because of the increasing need for security in different public places, voice biometrics may be a good solution, as we can easily take voice records. This paper provides a brief overview of the approaches utilized in recognizing speakers, and then presents a novel approach for recognizing speakers in degraded smart-home conditions. The suggested approach includes a pre-processing phase, a feature extraction phase, and a classification phase, where the feature extraction phase consists of formant extraction to get the spectrum energy maxima of speech audio, dynamic time warping (DTW)to find an optimal alignment between two provided temporal sequences under definite restrictions, and refinement process to improve the results of the DTW system output. The experiments are carried out on a database containing 1,248 samples in order to validate the suggested approach. The latter has good results as regards the state of the art with 94.5% accuracy. © 2023 IEEE.

关键词： Feature extraction

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：