Article
Computer Science, Information Systems
Marwa Jabberi, Ali Wali, Bilel Neji, Taha Beyrouthy, Adel M. Alimi
Summary: In this paper, a deep learning-based method for 3D face recognition is proposed. The method does not rely on using face representation methods as a proxy step for Convolutional Neural Networks (CNNs). Instead, 3D ShapeNets are employed for recognizing faces covering the full 3D shape, along with 3D data augmentation techniques to enlarge datasets. The experimental results demonstrate significant improvement in 3D face recognition performance using deep 3D CNNs like 3D ShapeNets.
Article
Computer Science, Information Systems
Akshay Mool, J. Panda, Kapil Sharma
Summary: This paper presents a face detection algorithm that achieves faster results in high quality videos. The algorithm uses Convolutional-MTCNN as the base algorithm and solves the problem of occlusion by using probabilistic approaches.
MULTIMEDIA TOOLS AND APPLICATIONS
(2022)
Article
Computer Science, Information Systems
Marwa Jabberi, Ali Wali, Bidyut Baran Chaudhuri, Adel M. Alimi
Summary: This paper proposes a method for 3D face alignment of 2D face images in the wild with noisy landmarks. It reconstructs a 3D face model and performs alignment and pose correction using deep learning for face recognition. The proposed method achieves comparable or even better recognition performances compared to the best results reported on popular benchmarks.
MULTIMEDIA TOOLS AND APPLICATIONS
(2023)
Article
Engineering, Electrical & Electronic
Zhenyu Weng, Huiping Zhuang, Haizhou Li, Balakrishnan Ramalingam, Rajesh Elara Mohan, Zhiping Lin
Summary: This paper proposes a new online multi-face tracking method, OMTMCM, which improves tracking performance by utilizing both face and body information. The method consists of two stages: detection alignment and detection association. In the first stage, a detection alignment module is used to align face and body detections from the same person. In the second stage, a cascaded matching module associates face detections across frames using past face and body features. Experimental results show that OMTMCM performs on par with or better than other online tracking methods for multi-face tracking.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2023)
Article
Computer Science, Software Engineering
Diqiong Jiang, Yiwei Jin, Fang-Lue Zhang, Yu-Kun Lai, Risheng Deng, Ruofeng Tong, Min Tang
Summary: This study reconstructs distinctive 3D face shapes and introduces a novel shape identity-aware regularization (SIR) loss. By improving the discriminability in both shape parameter and shape geometry domains, our method achieves superior experimental results compared to existing methods.
COMPUTER GRAPHICS FORUM
(2022)
Article
Engineering, Electrical & Electronic
Meihua Xiao, Hanxiao Yi, Ying Huang, Guoliang Luo, Qian Xiao
Summary: In this work, a highly efficient 3D face point cloud optimization method based on Kinect is proposed, which improves the reconstruction quality by delicately applying diverse optimization strategies. Extensive experiments demonstrate that our approach outperforms traditional methods in terms of accuracy.
JOURNAL OF SENSORS
(2022)
Article
Computer Science, Information Systems
Ziqi Dong, Furong Tian, Hua Yang, Tao Sun, Wenchuan Zhang, Dan Ruan
Summary: Based on heterogeneous trajectories, our proposed framework matches face trajectories with corresponding mobile phone trajectories to achieve object tracking or trajectory prediction. Our solution consists of two stages: selecting phone trajectories for a given face trajectory and identifying which phone trajectory is an exact match. We use a Multi-Granularity SpatioTemporal Window Searching algorithm to select candidate mobile phones close to a given face, and then build an affinity function to score face-phone trajectory pairs and determine if they match. LightGBM achieves the best performance with 92.6% accuracy, 96.9% precision, 88.5% recall, and 92.5% F1. Our framework is applicable in most scenarios and may benefit downstream tasks.
Article
Computer Science, Software Engineering
Jiayi Xu, Xinying Xue, Yitiao Wu, Xiaoyang Mao
Summary: This paper focuses on matching a computer-generated composite face sketch to a photograph, proposing a robust feature model to blend different facial representation modalities for matching purposes. Experimental results show that this framework can achieve more satisfying results compared to existing methods by fusing features from various sources for matching.
Article
Psychology, Multidisciplinary
Xiaomei Zhou, Shruti Vyas, Jinbiao Ning, Margaret C. Moulson
Summary: The study compares the attentional mechanisms in adults and infants learning naturally varying faces, finding that infants have difficulty resisting contextual distractions during the learning process, leading to a potential lack of discrimination between familiar and novel faces.
PSYCHOLOGICAL SCIENCE
(2022)
Article
Computer Science, Information Systems
Changwei Luo, Juyong Zhang, Changcun Bao, Yali Li, Jing Huang, Shengjin Wang
Summary: This paper addresses the problem of robust 3D face modeling and tracking from RGB-D images. The proposed method estimates the initial head pose using random forests and fits a generic bilinear face model to the RGB-D image using the iterative closest point algorithm. It improves the accuracy and robustness by integrating optimal vertex weights and utilizing the distances between facial landmarks. The experimental results demonstrate that the method can generate accurate 3D face models and track the face robustly under large head rotations and various facial expressions.
MULTIMEDIA SYSTEMS
(2022)
Article
Computer Science, Information Systems
Seongmin Lee, Hyunse Yoon, Sohyun Park, Sanghoon Lee, Jiwoo Kang
Summary: This study introduces neural networks to reconstruct stable and precise 3D faces by learning facial changes caused by identity, expression, and temporal cues. The proposed facial alignment network exhibits reliable and precise performance in reconstructing static and dynamic faces.
Article
Computer Science, Interdisciplinary Applications
Ellorine Carle, Pascal Sirguey, Simon C. Cox
Summary: This study proposes and evaluates the performance of two image matching algorithms for measuring surface displacements on landslides using hillshades derived from high-resolution DSMs. The algorithms, including a widely used NCC algorithm and an optical flow approach, were applied to hillshades obtained from APM/SPM imagery captured in 2018 and 2020. The results demonstrate the effectiveness of the masking approach and the good performance of the optical flow algorithm in delivering dense, accurate displacement measurements, especially with high-resolution DSMs available.
COMPUTERS & GEOSCIENCES
(2023)
Article
Computer Science, Software Engineering
Zhengda Lu, Jianwei Guo, Jun Xiao, Ying Wang, Xiaopeng Zhang, Dong-Ming Yan
Summary: This paper introduces an automatic method based on quadric surface fitting technique for extracting complete feature curve networks (FCNs) and generating high-quality segmentation from 3D surface meshes. The algorithm is shown to be more robust for FCN extraction from complex input meshes and achieves higher quality patch layouts compared with existing approaches. The validity of extracted feature curve cycles is also verified by applying them to surface reconstruction.
COMPUTER-AIDED DESIGN
(2021)
Article
Behavioral Sciences
Paolo Masulli, Martyna Galazka, David Eberhard, Jakob Asberg Johnels, Christopher Gillberg, Eva Billstedt, Nouchine Hadjikhani, Tobias S. Andersen
Summary: Gaze patterns during face perception can predict autistic traits and depression symptoms. A data-driven method was used to analyze gaze patterns and their relation to diagnostic test scores, offering an alternative approach to gaze data analysis.
Article
Computer Science, Software Engineering
Xiuqiang Song, Weijian Xie, Jiachen Li, Nan Wang, Fan Zhong, Guofeng Zhang, Xueying Qin
Summary: In this study, a novel method for tracking object poses using rough models is proposed. The rough contour is reshaped through a probability map, and the inner region information of the object is emphasized. A pre-search for 2D translation is conducted to improve the initial pose. Experimental results demonstrate the excellent performance of the proposed method on both roughly and precisely modeled objects.
COMPUTER GRAPHICS FORUM
(2023)