Lighting Equilibrium Distribution Maps and Their Application to Face Recognition Under Difficult Lighting Conditions

2017 ◽  
Vol 31 (03) ◽  
pp. 1756003 ◽  
Author(s):  
Jun Dong ◽  
Xue Yuan ◽  
Fanlun Xiong
Keyword(s):  
Face Recognition ◽  
Equilibrium Distribution ◽  
Main Idea ◽  
Data Sets ◽  
Test Image ◽  
Distribution Maps ◽  
Lighting Conditions ◽  
Map Construction ◽  
Compensation Methods ◽  
Approximation Coefficients

In this paper, a novel facial-patch based recognition framework is proposed to deal with the problem of face recognition (FR) on the serious illumination condition. First, a novel lighting equilibrium distribution maps (LEDM) for illumination normalization is proposed. In LEDM, an image is analyzed in logarithm domain with wavelet transform, and the approximation coefficients of the image are mapped according to a reference-illumination map in order to normalize the distribution of illumination energy due to different lighting effects. Meanwhile, the detail coefficients are enhanced to achieve detail information emphasis. The LEDM is obtained by blurring the distances between the test image and the reference illumination map in the logarithm domain, which may express the entire distribution of illumination variations. Then, a facial-patch based framework and a credit degree based facial patches synthesizing algorithm are proposed. Each normalized face images is divided into several stacked patches. And, all patches are individually classified, then each patch from the test image casts a vote toward the parent image classification. A novel credit degree map is established based on the LEDM, which is deciding a credit degree for each facial patch. The main idea of credit degree map construction is the over-and under-illuminated regions should be assigned lower credit degree than well-illuminated regions. Finally, results are obtained by the credit degree based facial patches synthesizing. The proposed method provides state-of-the-art performance on three data sets that are widely used for testing FR under different illumination conditions: Extended Yale-B, CAS-PEAL-R1, and CMUPIE. Experimental results show that our FR frame outperforms several existing illumination compensation methods.

Anatomization of the systems of dimension relaxation for facial recognition

2020 ◽  
pp. 1-11
Author(s):  
Mayamin Hamid Raha ◽  
Tonmoay Deb ◽  
Mahieyin Rahmun ◽  
Tim Chen
Keyword(s):  
Face Recognition ◽  
Main Idea ◽  
Principal Component ◽  
Feature Space ◽  
Projection Methods ◽  
Data Representation ◽  
Linear Discriminant ◽  
Class Differences ◽  
Underlying Space ◽  
Maximum Variance

Face recognition is the most efficient image analysis application, and the reduction of dimensionality is an essential requirement. The curse of dimensionality occurs with the increase in dimensionality, the sample density decreases exponentially. Dimensionality Reduction is the process of taking into account the dimensionality of the feature space by obtaining a set of principal features. The purpose of this manuscript is to demonstrate a comparative study of Principal Component Analysis and Linear Discriminant Analysis methods which are two of the highly popular appearance-based face recognition projection methods. PCA creates a flat dimensional data representation that describes as much data variance as possible, while LDA finds the vectors that best discriminate between classes in the underlying space. The main idea of PCA is to transform high dimensional input space into the function space that displays the maximum variance. Traditional LDA feature selection is obtained by maximizing class differences and minimizing class distance.

Multi-Instance Dimensionality Reduction via Sparsity and Orthogonality

2018 ◽  
Vol 30 (12) ◽  
pp. 3281-3308
Author(s):  
Hong Zhu ◽  
Li-Zhi Liao ◽  
Michael K. Ng
Keyword(s):  
Dimensionality Reduction ◽  
Optimization Problem ◽  
Augmented Lagrangian ◽  
Main Idea ◽  
Real Data ◽  
Learning Performance ◽  
High Dimensional ◽  
Data Sets ◽  
Outer Loop ◽  
Orthogonality Constraints

We study a multi-instance (MI) learning dimensionality-reduction algorithm through sparsity and orthogonality, which is especially useful for high-dimensional MI data sets. We develop a novel algorithm to handle both sparsity and orthogonality constraints that existing methods do not handle well simultaneously. Our main idea is to formulate an optimization problem where the sparse term appears in the objective function and the orthogonality term is formed as a constraint. The resulting optimization problem can be solved by using approximate augmented Lagrangian iterations as the outer loop and inertial proximal alternating linearized minimization (iPALM) iterations as the inner loop. The main advantage of this method is that both sparsity and orthogonality can be satisfied in the proposed algorithm. We show the global convergence of the proposed iterative algorithm. We also demonstrate that the proposed algorithm can achieve high sparsity and orthogonality requirements, which are very important for dimensionality reduction. Experimental results on both synthetic and real data sets show that the proposed algorithm can obtain learning performance comparable to that of other tested MI learning algorithms.

Deep Learning for Biometric Face Recognition: Experimental Study on Benchmark Data Sets

2020 ◽  
pp. 71-97
Author(s):  
Natalya Selitskaya ◽  
S. Sielicki ◽  
L. Jakaite ◽  
V. Schetinin ◽  
F. Evans ◽  
...  
Keyword(s):  
Experimental Study ◽  
Deep Learning ◽  
Face Recognition ◽  
Data Sets ◽  
Benchmark Data

scholarly journals Implementation and Application of Fall Detection on Mobile Perception Robot

2021 ◽  
Vol 2136 (1) ◽  
pp. 012053
Author(s):  
Zeyu Chen
Keyword(s):  
Elderly Population ◽  
Fall Detection ◽  
The Elderly ◽  
Detection Algorithm ◽  
Data Sets ◽  
Construction Technology ◽  
Robot Operating System ◽  
Map Construction ◽  
Robot Perception ◽  
Window Method

Abstract With the rapid increase in the number of people living in the elderly population, reducing and dealing with the problem of falls in the elderly has become the focus of research for decades. It is impossible to completely eliminate falls in daily life and activities. Detecting a fall in time can protect the elderly from injury as much as possible. This article uses the Turtlebot robot and the ROS robot operating system, combined with simultaneous positioning and map construction technology, Monte Carlo positioning, A* path planning, dynamic window method, and indoor map navigation. The YOLO network is trained using the stance and fall data sets, and the YOLOv4 target detection algorithm is combined with the robot perception algorithm to finally achieve fall detection on the turtlebot robot, and use the average precision, precision, recall and other indicators to measure.

scholarly journals Implementing CCTV-Based Attendance Taking Support System Using Deep Face Recognition: A Case Study at FPT Polytechnic College

Symmetry ◽  
2020 ◽  
Vol 12 (2) ◽  
pp. 307 ◽  
Author(s):  
Ngo Tung Son ◽  
Bui Ngoc Anh ◽  
Tran Quy Ban ◽  
Le Phuong Chi ◽  
Bui Dinh Chien ◽  
...  
Keyword(s):  
Face Recognition ◽  
Motion Blur ◽  
Security Monitoring ◽  
Flexible System ◽  
Lighting Conditions ◽  
Satisfactory Performance ◽  
Measurement Results ◽  
The Face ◽  
Recognition Systems

Face recognition (FR) has received considerable attention in the field of security, especially in the use of closed-circuit television (CCTV) cameras in security monitoring. Although significant advances in the field of computer vision are made, advanced face recognition systems provide satisfactory performance only in controlled conditions. They deteriorate significantly in the face of real-world scenarios such as lighting conditions, motion blur, camera resolution, etc. This article shows how we design, implement, and conduct the empirical comparisons of machine learning open libraries in building attendance taking (AT) support systems using indoor security cameras called ATSS. Our trial system was deployed to record the appearances of 120 students in five classes who study on the third floor of FPT Polytechnic College building. Our design allows for flexible system scaling, and it is not only usable for a school but a generic attendance system with CCTV. The measurement results show that the accuracy is suitable for many different environments.

scholarly journals P-wave first-motion polarity determination of waveform data in western Japan using deep learning

2019 ◽  
Vol 71 (1) ◽  
Author(s):  
Shota Hara ◽  
Yukitoshi Fukahata ◽  
Yoshihisa Iio
Keyword(s):  
Training Data ◽  
P Wave ◽  
Data Sets ◽  
Waveform Data ◽  
Arrival Times ◽  
Distribution Maps ◽  
Seismic Waveforms ◽  
Polarity Determination ◽  
Western Japan ◽  
First Motion

AbstractP-wave first-motion polarity is the most useful information in determining the focal mechanisms of earthquakes, particularly for smaller earthquakes. Algorithms have been developed to automatically determine P-wave first-motion polarity, but the performance level of the conventional algorithms remains lower than that of human experts. In this study, we develop a model of the convolutional neural networks (CNNs) to determine the P-wave first-motion polarity of observed seismic waveforms under the condition that P-wave arrival times determined by human experts are known in advance. In training and testing the CNN model, we use about 130 thousand 250 Hz and about 40 thousand 100 Hz waveform data observed in the San-in and the northern Kinki regions, western Japan, where three to four times larger number of waveform data were obtained in the former region than in the latter. First, we train the CNN models using 250 Hz and 100 Hz waveform data, respectively, from both regions. The accuracies of the CNN models are 97.9% for the 250 Hz data and 95.4% for the 100 Hz data. Next, to examine the regional dependence, we divide the waveform data sets according to the observation region, and then we train new CNN models with the data from one region and test them using the data from the other region. We find that the accuracy is generally high ($${ \gtrsim }$$≳ 95%) and the regional dependence is within about 2%. This suggests that there is almost no need to retrain the CNN model by regions. We also find that the accuracy is significantly lower when the number of training data is less than 10 thousand, and that the performance of the CNN models is a few percentage points higher when using 250 Hz data compared to 100 Hz data. Distribution maps, on which polarities determined by human experts and the CNN models are plotted, suggest that the performance of the CNN models is better than that of human experts.

Sign in / Sign up

Export Citation Format

Share Document