Deeply Learned Pose Invariant Image Analysis with Applications in 3D Face Recognition

Face alignment and face recognition are two fundamental problems in the facial analysis community. For face alignment, it forms the basis for the accurate face recognition, age estimation, and facial expression recognition. For face recognition, it has been widely applied in various practical scenarios such as access control system, massive surveillance, human computer interaction, etc.. There mainly exist two lines of works in these two fields, namely holistic face alignment and recognition, and partial face alignment and recognition. Numerous holistic face alignment and recognition works

A crucial part of the face recognition process, along with facial attribute analysis and face verification, is facial landmark detection. In order to prepare face images for feature extraction in facial analysis jobs, face alignment is an essential step. Alignment is frequently used in training and inference to standardize the locations of important facial landmarks for applications including face recognition, facial emotion detection, and facial attribute classification. The effectiveness of facial analysis models is known to be greatly impacted by the approach and application of face alignment. However, there has not been enough research done on how alignment affects the quality of facial images. Although face alignment techniques based on convolutional neural networks have advanced significantly recently, occlusion remains a fundamental obstacle to achieving high accuracy. In order to enhance performance, we present in this study the attentioned distillation module from our earlier work on the Occlusion-Adaptive Deep Network model. A distillation module in this model infers the occlusion probability of every place in high-level features. When measuring the relationship between facial shape and look, it can be automatically learned. However, because of the missing semantic features, the holistic face cannot be represented by the clean feature representation. We must use a low-rank learning module to recover lost features in order to achieve exhaustive and complete feature representation.

Unlike 2D face recognition (FR), the problem of insufficient training data is a major difficulty in 3D face recognition. Traditional Convolutional neural networks (CNNs) can not comprehensively learn all proper filters for FR applications. We embed a handcrafted feature map into our CNN framework—A hybrid data representation is proposed for 3D face. Furthermore, we use a Squeeze-Excitation block to learn the weights of data channels from training face datasets. To overcome the bias of training model based on a small 3D dataset, transfer learning is applied by fine-turning pre-training models, which is trained based on a large 2D face datasets. Tests show that, under challenge conditions such as expression and occlusion, our method outperforms other state-of-the-art methods and can run in real-time.

Over the years, face recognition has been the research topic that has attracted many researchers around the world. One of the most significant applications of face recognition is the access control system. The access control system allows authorized persons to enter or exit certain or restricted areas. As a result, it will increase the security situation without over-investment in staff security. The access information can be the identification, time, and location, etc. It can be used to carry out human resource management tasks such as attendance and inspection of employees in a more fair and transparent manner. Although face recognition has been widely used in access control systems because of its better accuracy and convenience without requiring too much user cooperation, the 2D-based face recognition systems also retain many limitations due to the variations in pose and illumination. By analyzing facial geometries, 3D facial recognition systems can theoretically overcome the disadvantages of prior 2D methods and improve robustness in different working conditions. In this paper, we propose the 3D facial recognition algorithm for use in an access control system. The proposed algorithm includes the preprocessing, feature extraction, and classification stages. The application of the proposed access control system is the automatic sliding door, the controller of the system, the web-based monitoring, control, and storage of data.

Face Recognition is an efficient technique and one of the most liked biometric software application for the identification and verification of specific individual in a digital image by analysing and comparing patterns. This paper presents a survey on well-known techniques of face recognition. The primary goal of this review is to observe the performance of different face recognition algorithms such as SVM (Support Vector Machine), CNN (Convolutional Neural Network), Eigenface based algorithm, Gabor Wavelet, PCA (Principle Component Analysis) and HMM (Hidden Markov Model). It presents comparative analysis about the efficiency of each algorithm. This paper also figure out about various face recognition applications used in real world and face recognition challenges like Illumination Variation, Pose Variation, Occlusion, Expressions Variation, Low Resolution and Ageing in brief. Another interesting component covered in this paper is review of datasets available for face recognition. So, must needed survey of many recently introduced face recognition aspects and algorithms are presented.

In this paper, we present a novel pose and expression invariant 3D face recognition approach based on intrinsic coordinate system (ICS) alignment. Motivated by the fact that a single classifier cannot be generally efficient against all face regions, a two tier region ensemble based classification approach is presented which employs three parallel face recognition algorithms using Mahalanobis Cosine (MahCos) matching score. The parallel face recognition algorithms employ Principal Component Analysis (PCA) based holistic features, Local Binary Patterns (LBP) based local features and modified Borda Count (MBC) based fusion technique respectively to classify the face images. The results obtained from the parallel algorithms are combined using an exponential rank reordering approach. The performance of the proposed methodology is corroborated by extensive experiments performed on FRGC v2.0 3D database. The results confirm that fusion strategies can be effectively used to construct a single classifier for improved performance.

Application of Face Recognition Method Under Deep Learning Algorithm in Embedded Systems

Machine face recognition has traditionally been studied under the assumption of a carefully controlled image acquisition process. By controlling image acquisition, variation due to factors such as pose, lighting, and background can be either largely eliminated or specifically limited to a study over a discrete number of possibilities. Applications of face recognition have had mixed success when deployed in conditions where the assumption of controlled image acquisition no longer holds. This dissertation focuses on this unconstrained face recognition problem, where face images exhibit the same amount of variability that one would encounter in everyday life. We formalize unconstrained face recognition as a binary pair matching problem (verification), and present a data set for benchmarking performance on the unconstrained face verification task. We observe that it is comparatively much easier to obtain many examples of unlabeled face images than face images that have been labeled with identity or other higher level information, such as the position of the eyes and other facial features. We thus focus on improving unconstrained face verification by leveraging the information present in this source of weakly supervised data. We first show how unlabeled face images can be used to perform unsupervised face alignment, thereby reducing variability in pose and improving verification accuracy. Next, we demonstrate how deep learning can be used to perform unsupervised feature discovery, providing additional image representations that can be combined with representations from standard hand-crafted image descriptors, to further improve recognition performance. Finally, we combine unsupervised feature learning with joint face alignment, leading to an unsupervised alignment system that achieves gains in recognition performance matching that achieved by supervised alignment.

With the development of deep learning, the accuracy of face recognition has been significantly improved. Current face recognition systems are mostly designed for CPU or GPU platforms, and faces significant latency and power constraints when migrated to embedded devices. In this live demonstration, a real-time face recognition system based on FPGA System-on-Chip (SoC) platforms is presented. To achieve real-time processing, the face recognition algorithm based on convolutional neural network is optimized first to a hardware-friendly network model and is accelerated on FPGA, while the face detection and face alignment are implemented on ARM. The latency of the entire system is 52 ms, and the face recognition accuracy on the LWF data set reaches 99.05%.

Recently, face recognition became a key element in social cognition which is used in various applications including human–robot interaction (HRI), pedestrian identification, and surveillance systems. Deep convolutional neural networks (CNNs) have achieved notable progress in recognizing faces. However, achieving accurate and real-time face recognition is still a challenging problem, especially in unconstrained environments due to occlusion, lighting conditions, and the diversity in head poses. In this paper, we present a robust face recognition and tracking framework in unconstrained settings. We developed our framework based on lightweight CNNs for all face recognition stages, including face detection, alignment and feature extraction, to achieve higher accuracies in these challenging circumstances while maintaining the real-time capabilities required for HRI systems. To maintain the accuracy, a single-shot multi-level face localization in the wild (RetinaFace) is utilized for face detection, and additive angular margin loss (ArcFace) is employed for recognition. For further enhancement, we introduce a face tracking algorithm that combines the information from tracked faces with the recognized identity to use in the further frames. This tracking algorithm improves the overall processing time and accuracy. The proposed system performance is tested in real-time experiments applied in an HRI study. Our proposed framework achieves real-time capabilities with an average of 99%, 95%, and 97% precision, recall, and F-score respectively. In addition, we implemented our system as a modular ROS package that makes it straightforward for integration in different real-world HRI systems.

Deep learning, in particular Convolutional Neural Network (CNN), has achieved promising results in face recognition recently. However, it remains an open question: why CNNs work well and how to design a 'good' architecture. The existing works tend to focus on reporting CNN architectures that work well for face recognition rather than investigate the reason. In this work, we conduct an extensive evaluation of CNN-based face recognition systems (CNN-FRS) on a common ground to make our work easily reproducible. Specifically, we use public database LFW (Labeled Faces in the Wild) to train CNNs, unlike most existing CNNs trained on private databases. We propose three CNN architectures which are the first reported architectures trained using LFW data. This paper quantitatively compares the architectures of CNNs and evaluate the effect of different implementation choices. We identify several useful properties of CNN-FRS. For instance, the dimensionality of the learned features can be significantly reduced without adverse effect on face recognition accuracy. In addition, traditional metric learning method exploiting CNN-learned features is evaluated. Experiments show two crucial factors to good CNN-FRS performance are the fusion of multiple CNNs and metric learning. To make our work reproducible, source code and models will be made publicly available.

Existing methods for iris, fingerprint, and 3D face recognition in mobile devices have constraints in terms of price and size owing to their use of additional cameras, lighting, and sensors. Additionally, visible light, camera-based 2D face recognition, palm print recognition, touchless fingerprint recognition, and finger knuckle print recognition are difficult to be used in mobile devices due to limitations in recognition performance and user inconvenience. In response to these problems, studies have been conducted on finger wrinkle recognition in mobile devices; however, image quality is often reduced by motion blurring caused by the movement of the camera or the user's finger, thereby reducing recognition performance. This study proposes a method for restoring and recognizing motion-blurred finger wrinkle images based on a generative adversarial network and deep convolutional neural network. Experiments were performed using two types of finger wrinkle databases, which were custom-made from images of 33 people captured by smart phone cameras (Dongguk mobile finger wrinkle database versions 1 and 2, denoted as DMFW-DB1 and DMFW-DB2, respectively). The results demonstrated high restoration and recognition performance in comparison with the state-of-the-art methods.

In this paper, a noble nonintrusive three-dimensional (3D) face modeling system for random-profile-based 3D face recognition is presented. Although recent two-dimensional (2D) face recognition systems can achieve a reliable recognition rate under certain conditions, their performance is limited by internal and external changes, such as illumination and pose variation. To address these issues, 3D face recognition, which uses 3D face data, has recently received much attention. However, the performance of 3D face recognition highly depends on the precision of acquired 3D face data, while also requiring more computational power and storage capacity than 2D face recognition systems. In this paper, we present a developed nonintrusive 3D face modeling system composed of a stereo vision system and an invisible near-infrared line laser, which can be directly applied to profile-based 3D face recognition. We further propose a novel random-profile-based 3D face recognition method that is memory-efficient and pose-invariant. The experimental results demonstrate that the reconstructed 3D face data consists of more than 50 k 3D point clouds and a reliable recognition rate against pose variation.

3D face modeling from 2D face images is of significant importance for face analysis, animation and recognition. Previous research on this topic mainly focused on 3D face modeling from a single 2D face image; however, a single face image can only provide a limited description of a 3D face. In many applications, for example, law enforcement, multi-view face images are usually captured for a subject during enrollment, which makes it desirable to build a 3D face texture model, given a pair of frontal and profile face images. We first determine the correspondence between un-calibrated frontal and profile face images through facial landmark alignment. An initial 3D face shape is then reconstructed from the frontal face image, followed by shape refinement utilizing the depth information provided by the profile image. Finally, face texture is extracted by mapping the frontal face image on the recovered 3D face shape. The proposed method is utilized for 2D face recognition in two scenarios: (i) normalization of probe image, and (ii) enhancing the representation capability of gallery set. Experimental results comparing the proposed method with a state-of-the-art commercial face matcher and densely sampled LBP on a subset of the FERET database show the effectiveness of the proposed 3D face texture model.

Face recognition is the essential security system, which is subjected to get more scrutiny in recent years especially in the field of research and also in industry. This study addresses the various approaches for recognizing face back on neural network by adopting convolutional neural network (CNN). The study has done on different techniques of face alignment, preprocessing techniques, and also in the size of the face images. This paper explains the computational analysis of face recognition system and emphasizes the accuracies and constraints of the images. The predominant face alignment approaches used are Dlib and constrained local model (CLM). For training, Tan-Triggs preprocessing technique is used in face image size of 96 × 96 and 64 × 64. The face recognition grand challenge (FRGC) dataset is used for the analysis, and it produced the accuracy of range from 90 to 98.30% on corresponding approaches.