3D Face Database Research Articles

Depth and Intensity Gabor Features Based 3D Face Recognition Using Symbolic LDA and AdaBoost

In this paper, the objective is to investigate what contributions depth and intensity informat ion make to the solution of face recognition problem when expression and pose variations are taken into account, and a novel system is proposed for combin ing depth and intensity information in order to improve face recognition performance. In the proposed approach, local features based on Gabor wavelets are extracted fro m depth and intensity images, wh ich are obtained fro m 3D data after fine align ment. Then a novel hierarch ical selecting scheme embedded in symbolic linear discriminant analysis (Symbolic LDA) with AdaBoost learning is proposed to select the most effective and robust features and to construct a strong classifier. Experiments are performed on the three datasets, namely, Texas 3D face database, Bhosphorus 3D face database and CASIA 3D face database, which contain face images with comp lex variations, including expressions, poses and longtime lapses between two scans. The experimental results demonstrate the enhanced effectiveness in the performance of the proposed method. Since most of the design processes are performed automatically, the proposed approach leads to a potential prototype design of an automat ic face recognition system based on the combination o f the depth and intensity informat ion in face images.

3D Face Recognition Based on Depth and Intensity Gabor Features using Symbolic PCA and AdaBoost

Traditional 2D face recognition based on optical (intensity or color) images faces many challenges, such as illumination, expression, and pose variation. In fact, the human face generates not only 2D texture information but also 3D shape information. In this paper, the objective is to investigate what contributions depth and intensity information make to the solution of face recognition problem when expression and pose variations are taken into account, and a novel system is proposed for combining depth and intensity information in order to improve face recognition performance. In the proposed approach, local features based on Gabor wavelets are extracted from depth and intensity images, which are obtained from 3D data after fine alignment. Then a novel hierarchical selecting scheme embedded in symbolic principal component analysis (Symbolic PCA) and AdaBoost learning is proposed to select the most effective and most robust features and to construct a strong classifier. Experiments are performed on the three datasets, namely,Texas 3D face database, Bhosphorus 3D face database and CASIA 3D face database, which contain face images with complex variations, including expressions, poses and long time lapses between two scans. The experimental results demonstrate the enhanced effectiveness in the performance of the proposed method. Since most of the design processes are performed automatically, the proposed approach leads to a potential prototype design of an automatic face recognition system based on the combination of the depth and intensity information in face images.

3D Facial Landmark Detection under Large Yaw and Expression Variations

A 3D landmark detection method for 3D facial scans is presented and thoroughly evaluated. The main contribution of the presented method is the automatic and pose-invariant detection of landmarks on 3D facial scans under large yaw variations (that often result in missing facial data), and its robustness against large facial expressions. Three-dimensional information is exploited by using 3D local shape descriptors to extract candidate landmark points. The shape descriptors include the shape index, a continuous map of principal curvature values of a 3D object's surface, and spin images, local descriptors of the object's 3D point distribution. The candidate landmarks are identified and labeled by matching them with a Facial Landmark Model (FLM) of facial anatomical landmarks. The presented method is extensively evaluated against a variety of 3D facial databases and achieves state-of-the-art accuracy (4.5-6.3 mm mean landmark localization error), considerably outperforming previous methods, even when tested with the most challenging data.

3D face alignment and registration in the presence of facial expression differences

AbstractThis paper deals with the problem of 3D alignment of faces in the presence of some facial expression changes. The data are three dimensional and obtained using a laser scanner. Our approach is based on the differential geometry of the surface and computes the intrinsic local fiducial points on the surface and on curves that reside on the surface. Because these fiducial points are local, they allow partial alignment, where part of the face is viewed. Moreover, since the fiducial points are relatively affine‐invariant to local affine transformations, they allow matching and alignment when facial expression changes affect part of the face. A fast, noniterative alignment procedure is presented in this paper that establishes reliable correspondences between fiducial points without any prior knowledge of the overall nonlinear global transformation that takes place after the changes in facial expressions. This is achieved through the construction of a set of ordered novel absolute local affine invariants. With enough fiducial points set as correspondents, the overall nonlinear transformation is computed and the face before and after the transformation are aligned. For comparison, we also compare the alignment performance of our method with that of the iterative closest point (ICP) method and the coherent point drift (CPD) method which is based on the Gaussian mixture model by looking at the between‐to‐within variation (signal‐to‐noise ratio, SNR) between these two classes (match vs nonmatch) based on the average alignment errors for the genuine matches and nonmatches, normalized by their respective within variation, and by running it on the 3D face database GavabDB. The separation between the true match and nonmatch is best for our zero‐torsion method (SNR of 1.2), followed by the ICP method (SNR of 0.37)and the CPD method (SNR of 0.15). It is interesting to observe here that, although the alignment error for the true match cases is lowest for the CPD method, it is also extremely low for the nonmatch cases, which would mean that the method would force a query into a wrong face, yielding poor specificity (lots of false alarms). © 2013 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.

Iterative Closest Normal Point for 3D Face Recognition

The common approach for 3D face recognition is to register a probe face to each of the gallery faces and then calculate the sum of the distances between their points. This approach is computationally expensive and sensitive to facial expression variation. In this paper, we introduce the iterative closest normal point method for finding the corresponding points between a generic reference face and every input face. The proposed correspondence finding method samples a set of points for each face, denoted as the closest normal points. These points are effectively aligned across all faces, enabling effective application of discriminant analysis methods for 3D face recognition. As a result, the expression variation problem is addressed by minimizing the within-class variability of the face samples while maximizing the between-class variability. As an important conclusion, we show that the surface normal vectors of the face at the sampled points contain more discriminatory information than the coordinates of the points. We have performed comprehensive experiments on the Face Recognition Grand Challenge database, which is presently the largest available 3D face database. We have achieved verification rates of 99.6 and 99.2 percent at a false acceptance rate of 0.1 percent for the all versus all and ROC III experiments, respectively, which, to the best of our knowledge, have seven and four times less error rates, respectively, compared to the best existing methods on this database.

基于Gabor特征核协作表达的三维人脸识别

To overcome the problem of illumination,facial expression and pose variants in 3D face recognition,the algorithm based on the Gabor features combined with kernel collaborative representation is proposed.Collaborative representation classification algorithm uses similar faces to describe testing faces collaboratively.Via solving sparse coefficient by l2 norm,collaborative representation classification algorithm can classify the testing face correctly according to reconstruction error. This method extracts 40 Gabor features with different scales and orientations of depth maps by the Gabor filter firstly.Secondly,these features are mapped to a high-dimensional space by selecting an appropriate kernel function. Finally,non-linear dimensionality reduction and feature selection fulfill the recognition task in high-dimensional space combining with collaborative representation classification.Extensive experiments on Kinect Face Dataset and Texas 3D face database demonstrate that the proposed algorithm is more effective than recently algorithms even when the number of training samples is small.

A robust invariant bipolar representation for R 3 surfaces: applied to the face description

In this paper, we intend to introduce a novel invariant curved surface representation under the 3D motion group. It is constructed from the superposition of the two geodesic potentials generated from a given couple of surface points. By sampling this continuous representation, invariant points are extracted from a large neighborhood around these reference points. Different numerical methods are implemented in order to find an efficient approximation in the mean of the shape distance. The inference of small distortions of points positions applied to the reference points is analyzed. We apply the proposed representation to real 3D images. The experimentations are performed on the 3D facial database Bosphorus.

P2-29: Comparing the Other-Race-Effect and Congenital Prosopagnosia Using a Three-Experiment Test Battery

Congenital prosopagnosia, an innate impairment in recognizing faces, as well as the other-race-effect, the disadvantage in recognizing faces of foreign races, both influence face recognition abilities. Here we compared both phenomena by testing three groups: German congenital prosopagnosics (cPs), unimpaired German, and unimpaired South Korean participants (n = 23 per group), on three tests with Caucasian faces. First we ran the Cambridge Face Memory Test (Duchaine & Nakayama, 2006 Neuropsychologia44 576–585). Participants had to recognize Caucasian target faces in a 3AFC task. German controls performed better than Koreans (p = .009) who performed better than prosopagnosics (p = .0001). Variation of the individual performances was larger for cPs than for Koreans (p = .028). In the second experiment, participants rated the similarity of Caucasian faces (in-house 3D face-database) which differed parametrically in features or second order relations (configuration). We found differences between sensitivities ...

A Supervised Hybrid Methodology for Pose and Illumination Invariant 3D Face Recognition

D face recognition systems encounter difficulties in recognizing faces with illumination variations. The depth map of the 3D face data has the potential to handle the variation in illumination of face images. The view variations are handled by using the moment invariants. Moment Invariants are used as rotation invariant features of the face image. For feature matching an efficient fuzzy-neural technique is proposed. The PCA components of normalized depth map and Moment invariants on mesh images are used successfully to implement a fuzzy neural network based fully automatic 3D face recognition system. The system is evaluated on the 3D face databases, the CASIA database. The proposed system provides recognition accuracy that is resulted in to an efficient 3D face recognition system. Keywords-Neural Network (FNN), 3D-Face Recognition, Depth Map, Moment Invariants.

Robust sparse bounding sphere for 3D face recognition

A robust sparse bounding sphere representation (RSBSR) is proposed to analyze 3D facial data. There are many obstacles to distinguishing facial differences, such as large pose and expression variations, hair occlusions and noise corruptions. In our framework, 3D point clouds are first preprocessed to remove the irrelevant areas and to align with a frontal neutral face model for overcoming the influence of large pose variations based on axis-angle representation. Then, 3D facial models are projected on the bounding spheres to describe both the depth and 3D geometric shape information, referred to as bounding sphere representation (BSR). This descriptor has the potential of decreasing the influence of large expression and pose variations on each normalized face within the corresponding spherical domain. Next, a robust group sparse regression model (RGSRM) is proposed to estimate the regression matrix, which preserves the intrinsic discriminant information. By embedding the descriptors into the low dimensional regression matrix, hair occlusions and artifacts can be treated as corruptions and can be patched. Under the constraints of Spectral Regression and corruptions, noise corruptions can be removed and the remaining small variations can be further corrected. FRGC v2.0 and CASIA 3D face databases are introduced to examine the performance of our framework and the previous algorithms with different schemes, and the experimental results show our proposed framework has the performance of simple implementation, high accuracy and low computational complexity.

Expansion of 3D face sample set based on genetic algorithm

3D face database is an important data platform for model training and algorithm design. As these works are mainly based on statistical learning, the coverage of training set has a great impact on algorithm performance. However it is a tedious process to obtain a 3D face sample. So the capacities of current 3D face databases are relatively insufficient. To solve this problem, we present a framework to augment existing 3D databases based on Genetic Algorithm. First the prototypical face samples are divided into patches. Then the new face samples are generated by assembling randomly selected patches. Under the guidance of the genetic algorithm, we can perform a number of the generating works at a time. The experiment results show that the proposed method has good performance on face data expansion.

Virtual structured-light coding for three-dimensional shape data compression

In order to reduce the file size of three-dimensional (3D) shape data, we propose two compression algorithms: two-channel phase coding algorithm and three-channel phase coding algorithm. In these two algorithms, 3D shape information is encoded into the color channels of a single 24-bit color image based on virtual structured-light projection system. Tested with a hemisphere of 1mm diameter used in another virtual structured-light coding algorithm, which was proposed by Nikolaus Karpinsky and Song Zhang in ‘composite phase-shifting algorithm for three-dimensional shape compression’ in 2010, the two-channel phase coding algorithm achieves compression ratio 1:56.6 with reconstruction error of ±4.91×10−4mm, and the three-channel algorithm gains compression ratio 1:33.8 with reconstruction error of ±0.36×10−4mm. The theoretical analyses demonstrate that the relative reconstruction errors of our coding algorithms can be reduced to 7.66×10−6 and 2.99×10−8 of the height of the 3D object, respectively. With these theoretical analyses, the virtual structured-light coding algorithms can be used to achieve desired reconstruction qualities with high compression ratios in storing, transmitting, encrypting 3D shapes and constructing 3D face databases.

REPRESENTING 3D FACE FROM POINT CLOUD TO FACE-ALIGNED SPHERICAL DEPTH MAP

We propose a novel representation of 3D face shape which is a key step for feature extraction and face recognition. The input of the proposed methods is unstructured point cloud, which determines the wide applicability of the proposed representation. Our contributions mainly include two parts: Spherical Depth Map (SDM) and face alignment based on SDM. SDM, which can be adopted to many applications, is a special kind of range image utilizing the prior anatomical knowledge of human face. Useful characteristics of SDM facilitate face alignment with higher efficiency and accuracy. Experiments conducted on three popular 3D face databases verify the high efficacy and superiority of the proposed method. The accuracy of face alignment is up to 100% with our strategy. The face verification rates based on the standard protocols are all higher than the baseline performance of FRGC2.0.

3D Novel Face Model Based on Surface Stitching

3D face sample is an important data platform for model training, algorithm design. Subject to the constraint of data acquisition equipment the size of current 3D face databases are relatively small and insufficient. To solve this problem, this paper presents a modeling way for generating 3D novel samples based on surface stitching. First we use morphable model to build a global model. Then, we replace each patch of the global model based on surface stitching. We demonstrate that with appropriate choice of local models it is possible to reliably generate new realistic face samples.

3D Novel Face Sample Modeling for Face Recognition

In this paper, we present a method to generate novel realistic 3D face model using a model trained from real 3D face data. 3D face samples are an important data platform for model training, algorithm design. Subject to the constraint of acquisition technology, sizes of current 3D face databases are relatively small and insufficient. The presented method is used to solve this problem. First real 3D face data are collected as training sample and a parametric global model is learned based on them. Then a local model is established based on surface quilting. We use the global model to build a novel coarse face model. Then, we condition the local model with the global model. With appropriate choices of local and global models it is possible to reliably generate novel realistic 3D face data that do not correspond to any individual in the training data. Finally we apply our model to face recognition to examine the meaning of our work

Face Recognition with Discriminating 3D DCT Coefficients

In this study, we investigate the use of a 3D discrete cosine transform (DCT) for 3D face recognition and present a novel 3D DCT-based feature extraction method with the selection of discriminating coefficients. We apply a 3D DCT on the voxel data, and use transform coefficients as features. Then the most discriminating 3D transform coefficients are selected with the proportion of variance, sequential floating forward selection and sequential floating backward selection methods. After feature selection, the linear discriminant analysis is applied on reduced sized feature vectors. We compare the results of different feature selection methods and show that a hybrid feature selection method has the best performance both in terms of time and recognition. Our experimental results verify that the discriminating DCT coefficients increase the face recognition rate more than the low-indexed coefficients do. On the other hand, the discriminating coefficients have only an energy level of 1.58%, too low when compared with the total energy of low-indexed coefficients. This fact shows that the discriminating coefficients are not the most energetic ones. With these coefficients, a recognition rate of 99.25% is achieved and this result is compared with other methods tested on a 3D RMA face database.

IMAGE BASED 3D FACE RECONSTRUCTION: A SURVEY

The use of 3D data in face image processing applications has received considerable attention during the last few years. A major issue for the implementation of 3D face processing systems is the accurate and real time acquisition of 3D faces using low cost equipment. In this paper we provide a survey of 3D reconstruction methods used for generating the 3D appearance of a face using either a single or multiple 2D images captured with ordinary equipment such as digital cameras and camcorders. In this context we discuss various issues pertaining to the general problem of 3D face reconstruction such as the existence of suitable 3D face databases, correspondence of 3D faces, feature detection, deformable 3D models and typical assumptions used during the reconstruction process. Different approaches to the problem of 3D reconstruction are presented and for each category the most important advantages and disadvantages are outlined. In particular we describe example-based methods, stereo methods, video-based methods and silhouette-based methods. The issue of performance evaluation of 3D face reconstruction algorithms, the state of the art and future trends are also discussed.

Automatic 3D face recognition from depth and intensity Gabor features

As is well known, traditional 2D face recognition based on optical (intensity or color) images faces many challenges, such as illumination, expression, and pose variation. In fact, the human face generates not only 2D texture information but also 3D shape information. In this paper, we investigate what contributions depth and intensity information makes to face recognition when expression and pose variations are taken into account, and we propose a novel system for combining depth and intensity information to improve face recognition systems. In our system, local features described by Gabor wavelets are extracted from depth and intensity images, which are obtained from 3D data after fine alignment. Then a novel hierarchical selecting scheme embedded in linear discriminant analysis (LDA) and AdaBoost learning is proposed to select the most effective and most robust features and to construct a strong classifier. Experiments are performed on the CASIA 3D face database and the FRGC V2.0 database, two data sets with complex variations, including expressions, poses and long time lapses between two scans. Experimental results demonstrate the promising performance of the proposed method. In our system, all processes are performed automatically, thus providing a prototype of automatic face recognition combining depth and intensity information.

Deformation Modeling for Robust 3D Face Matching

Face recognition based on 3D surface matching is promising for overcoming some of the limitations of current 2D image-based face recognition systems. The 3D shape is generally invariant to the pose and lighting changes, but not invariant to the non-rigid facial movement, such as expressions. Collecting and storing multiple templates to account for various expressions for each subject in a large database is not practical. We propose a facial surface modeling and matching scheme to match 2.5D facial scans in the presence of both non-rigid deformations and pose changes (multiview) to a 3D face template. A hierarchical geodesic-based resampling approach is applied to extract landmarks for modeling facial surface deformations. We are able to synthesize the deformation learned from a small group of subjects (control group) onto a 3D neutral model (not in the control group), resulting in a deformed template. A user-specific (3D) deformable model is built by combining the templates with synthesized deformations. The matching distance is computed by fitting this generative deformable model to a test scan. A fully automatic and prototypic 3D face matching system has been developed. Experimental results demonstrate that the proposed deformation modeling scheme increases the 3D face matching accuracy.

Quasi-Facial Communication for Online Learning Using 3D Modeling Techniques

Online interaction with 3D facial animation is an alternative way of face-to-face communication for distance education. 3D facial modeling is essential for virtual educational environments establishment. This article presents a novel 3D facial modeling solution that facilitates quasi-facial communication for online learning. Our algorithm builds 3D facial models from a single image, with support of a 3D face database. First from the image, we extract a set of feature points, which are then used to automatically estimate the head pose parameters using the 3D mean face in our database as a reference model. After the pose recovery, a similarity measurement function is proposed to locate the neighborhood for the given image in the 3D face database. The scope of neighborhood can be determined adaptively using our cross-validation algorithm. Furthermore, the individual 3D shape is synthesized by neighborhood interpolation. Texture mapping is achieved based on feature points. The experimental results show that our algorithm can robustly produce 3D facial models from images captured in various scenarios to enhance the lifelikeness in distant learning.