Discriminative extended canonical correlation analysis for pattern set matching

In this paper we address the problem of matching sets of vectors embedded in the same input space. We propose an approach which is motivated by canonical correlation analysis (CCA), a statistical technique which has proven successful in a wide variety of pattern recognition problems. Like CCA when applied to the matching of sets, our extended canonical correlation analysis (E-CCA) aims to extract the most similar modes of variability within two sets. Our first major contribution is the formulation of a principled framework for robust inference of such modes from data in the presence of uncertainty associated with noise and sampling randomness. E-CCA retains the efficiency and closed form computability of CCA, but unlike it, does not possess free parameters which cannot be inferred directly from data (inherent data dimensionality, and the number of canonical correlations used for set similarity computation). Our second major contribution is to show that in contrast to CCA, E-CCA is readily adapted to match sets in a discriminative learning scheme which we call discriminative extended canonical correlation analysis (DE-CCA). Theoretical contributions of this paper are followed by an empirical evaluation of its premises on the task of face recognition from sets of rasterized appearance images. The results demonstrate that our approach, E-CCA, already outperforms both CCA and its quasi-discriminative counterpart constrained CCA (C-CCA), for all values of their free parameters. An even greater improvement is achieved with the discriminative variant, DE-CCA.     

Automatic segmentation of granular objects in images: Combining local density clustering and gradient-barrier watershed

Blob or granular object recognition is an image processing task with a rich application background, ranging from cell/nuclei segmentation in biology to nanoparticle recognition in physics. In this study, we establish a new and comprehensive framework for granular object recognition. Local density clustering and connected component analysis constitute the first stage. To separate overlapping objects, we further propose a modified watershed approach called the gradient-barrier watershed, which better incorporates intensity gradient information into the geometrical watershed framework. We also revise the marker-finding procedure to incorporate a clustering step on all the markers initially found, potentially grouping multiple markers within the same object. The gradient-barrier watershed is then conducted based on those markers, and the intensity gradient in the image directly guides the water flow during the flooding process. We also propose an important scheme for edge detection and fore/background separation called the intensity moment approach. Experimental results for a wide variety of objects in different disciplines – including cell/nuclei images, biological colony images, and nanoparticle images – demonstrate the effectiveness of the proposed framework.     

Face recognition using spectral features

Face recognition is a challenging task in computer vision and pattern recognition. It is well-known that obtaining a low-dimensional feature representation with enhanced discriminatory power is of paramount importance to face recognition. Moreover, recent research has shown that the face images reside on a possibly nonlinear manifold. Thus, how to effectively exploit the hidden structure is a key problem that significantly affects the recognition results. In this paper, we propose a new unsupervised nonlinear feature extraction method called spectral feature analysis (SFA). The main advantages of SFA over traditional feature extraction methods are: (1) SFA does not suffer from the small-sample-size problem; (2) SFA can extract discriminatory information from the data, and we show that linear discriminant analysis can be subsumed under the SFA framework; (3) SFA can effectively discover the nonlinear structure hidden in the data. These appealing properties make SFA very suitable for face recognition tasks. Experimental results on three benchmark face databases illustrate the superiority of SFA over traditional methods.     

Face recognition using laplacianfaces

We propose an appearance-based face recognition method called the Laplacianface approach. By using locality preserving projections (LPP), the face images are mapped into a face subspace for analysis. Different from principal component analysis (PCA) and linear discriminant analysis (LDA) which effectively see only the Euclidean structure of face space, LPP finds an embedding that preserves local information, and obtains a face subspace that best detects the essential face manifold structure. The Laplacianfaces are the optimal linear approximations to the eigenfunctions of the Laplace Beltrami operator on the face manifold. In this way, the unwanted variations resulting from changes in lighting, facial expression, and pose may be eliminated or reduced. Theoretical analysis shows that PCA, LDA, and LPP can be obtained from different graph models. We compare the proposed Laplacianface approach with Eigenface and Fisherface methods on three different face data sets. Experimental results suggest that the proposed Laplacianface approach provides a better representation and achieves lower error rates in face recognition.     

Discriminative sparse representation for face recognition

Recently Sparse Representation (or coding) based Classification (SRC) has gained great success in face recognition. In SRC, the testing image is expected to be best represented as a sparse linear combination of training images from the same class, and the representation fidelity is measured by the ?₂-norm or ?₁-norm of the coding residual. However, SRC emphasizes the sparsity too much and overlooks the spatial information during local feature encoding process which has been demonstrated to be critical in real-world face recognition problems. Besides, some work considers the spatial information but overlooks the different discriminative ability in different face regions. In this paper, we propose to weight spatial locations based on their discriminative abilities in sparse coding for robust face recognition. Specifically, we learn the weights at face locations according to the information entropy in each face region, so as to highlight locations in face images that are important for classification. Furthermore, in order to construct a robust weights to fully exploit structure information of each face region, we employed external data to learn the weights, which can cover all possible face image variants of different persons, so the robustness of obtained weights can be guaranteed. Finally, we consider the group structure of training images (i.e. those from the same subject) and added an ?_2,1-norm (group Lasso) constraint upon the formulation, which enforcing the sparsity at the group level. Extensive experiments on three benchmark face datasets demonstrate that our proposed method is much more robust and effective than baseline methods in dealing with face occlusion, corruption, lighting and expression changes, etc.     

Improved discriminant locality preserving projections for face and palmprint recognition

We propose in this paper two improved manifold learning methods called diagonal discriminant locality preserving projections (Dia-DLPP) and weighted two-dimensional discriminant locality preserving projections (W2D-DLPP) for face and palmprint recognition. Motivated by the fact that diagonal images outperform the original images for conventional two-dimensional (2D) subspace learning methods such as 2D principal component analysis (2DPCA) and 2D linear discriminant analysis (2DLDA), we first propose applying diagonal images to a recently proposed 2D discriminant locality preserving projections (2D-DLPP) algorithm, and formulate the Dia-DLPP method for feature extraction of face and palmprint images. Moreover, we show that transforming an image to a diagonal image is equivalent to assigning an appropriate weight to each pixel of the original image to emphasize its different importance for recognition, which provides the rationale and superiority of using diagonal images for 2D subspace learning. Inspired by this finding, we further propose a new discriminant weighted method to explicitly calculate the discriminative score of each pixel within a face and palmprint sample to duly emphasize its different importance, and incorporate it into 2D-DLPP to formulate the W2D-DLPP method to improve the recognition performance of 2D-DLPP and Dia-DLPP. Experimental results on the widely used FERET face and PolyU palmprint databases demonstrate the efficacy of the proposed methods.     

Face recognition using a kernel fractional-step discriminant analysis algorithm

Feature extraction is among the most important problems in face recognition systems. In this paper, we propose an enhanced kernel discriminant analysis (KDA) algorithm called kernel fractional-step discriminant analysis (KFDA) for nonlinear feature extraction and dimensionality reduction. Not only can this new algorithm, like other kernel methods, deal with nonlinearity required for many face recognition tasks, it can also outperform traditional KDA algorithms in resisting the adverse effects due to outlier classes. Moreover, to further strengthen the overall performance of KDA algorithms for face recognition, we propose two new kernel functions: cosine fractional-power polynomial kernel and non-normal Gaussian RBF kernel. We perform extensive comparative studies based on the YaleB and FERET face databases. Experimental results show that our KFDA algorithm outperforms traditional kernel principal component analysis (KPCA) and KDA algorithms. Moreover, further improvement can be obtained when the two new kernel functions are used.     

Subspace methods for face recognition

Studying the inherently high-dimensional nature of the data in a lower dimensional manifold has become common in recent years. This is generally known as dimensionality reduction. A very interesting strategy for dimensionality reduction is what is known as subspace analysis. Beginning with the Eigenface method, face recognition and in general computer vision has witnessed a growing interest in algorithms that capitalize on this idea and an ample number of such efficient algorithms have been proposed. These algorithms mainly differ in the kind of projection method used (linear or non-linear) or in the criterion employed for classification. The objective of this paper is to provide a comprehensive performance evaluation of about twenty five different subspace algorithms under several important real time test conditions. For this purpose, we have considered the performance of these algorithms on data taken from four standard face and object databases namely ORL, Yale, FERET and the COIL-20 object database. This paper also presents some theoretical aspects of the algorithm and the analysis of the simulations carried out.     

Real-Time Pose-Invariant Face Recognition Using the Efficient Second-Order Minimization and the Pose Transforming Matrix

We propose a real-time pose-invariant face recognition algorithm from a gallery of frontal images only. (i) We modified the second-order minimization method for the active appearance model (AAM). This allows the AAM to have the ability of correct convergence with little loss of frame rate. (ii) We proposed a pose transforming matrix that can eliminate warping artifacts of the warped face image from AAM fitting. This makes it possible to train a neural network as the face recognizer with one frontal face image of each person in the gallery set. (iii) We propose a simple method for pose recognition by using neural networks to select the proper pose transforming matrix. The proposed algorithm was evaluated on a set of 2000 facial images of 10 people (200 images for each person obtained at various poses), achieving a great improvement in recognition.     

Subspace distance analysis with application to adaptive Bayesian algorithm for face recognition

We propose subspace distance measures to analyze the similarity between intrapersonal face subspaces, which characterize the variations between face images of the same individual. We call the conventional intrapersonal subspace average intrapersonal subspace (AIS) because the image differences often come from a large number of persons. An intrapersonal subspace is referred to as specific intrapersonal subspace (SIS) if the image differences are from just one person. We demonstrate that SIS varies significantly from person to person, and most SISs are not similar to AIS. Based on these observations, we introduce the maximum a posteriori (MAP) adaptation to the problem of SIS estimation, and apply it to the Bayesian face recognition algorithm. Experimental results show that the adaptive Bayesian algorithm outperforms the non-adaptive Bayesian algorithm as well as Eigenface and Fisherface methods if a small number of adaptation images are available.     

Intrapersonal subspace analysis with application to adaptive Bayesian face recognition

We propose a subspace distance measure to analyze the similarity between intrapersonal face subspaces, which characterize the variations between face images of the same individual. We call the conventional intrapersonal subspace the average intrapersonal subspace (AIS) because the image differences often come from a large number of persons. We call an intrapersonal subspace specific intrapersonal subspace (SIS) if the image differences are from just one person. We demonstrate that SIS varies from person to person and most SISs are not similar to AIS. Based on these observations, we introduce the maximum a posteriori (MAP) adaptation to the problem of SIS estimation, and apply it to the Bayesian face recognition algorithm. Experimental results show that the adaptive Bayesian algorithm outperforms the non-adaptive Bayesian algorithm as well as Eigenface and Fisherface methods when a small number of adaptation images are available.     

3D face verification across pose based on euler rotation and tensors

In this paper, we propose a new approach for 3D face verification based on tensor representation. Face challenges, such as illumination, expression and pose, are modeled as a multilinear algebra problem where facial images are represented as high order tensors. Particularly, to account for head pose variations, several pose scans are generated from a single depth image using Euler transformation. Multi-bloc local phase quantization (MB-LPQ) histogram features are extracted from depth face images and arranged as a third order tensor. The dimensionality of the tensor is reduced based on the higher-order singular value decomposition (HOSVD). HOSVD projects the input tensor in a new subspace in which the dimension of each tensor mode is reduced. To discriminate faces of different persons, we utilize the Enhanced Fisher Model (EFM). Experimental evaluations on CASIA-3D database, which contains large head pose variations, demonstrate the effectiveness of the proposed approach. A verification rate of 98.60% is obtained.     

Diagonal principal component analysis for face recognition

In this paper, a novel subspace method called diagonal principal component analysis (DiaPCA) is proposed for face recognition. In contrast to standard PCA, DiaPCA directly seeks the optimal projective vectors from diagonal face images without image-to-vector transformation. While in contrast to 2DPCA, DiaPCA reserves the correlations between variations of rows and those of columns of images. Experiments show that DiaPCA is much more accurate than both PCA and 2DPCA. Furthermore, it is shown that the accuracy can be further improved by combining DiaPCA with 2DPCA.     

Face recognition by sparse discriminant analysis via joint L2,1-norm minimization

Recently, joint feature selection and subspace learning, which can perform feature selection and subspace learning simultaneously, is proposed and has encouraging ability on face recognition. In the literature, a framework of utilizing L_2,1-norm penalty term has also been presented, but some important algorithms cannot be covered, such as Fisher Linear Discriminant Analysis and Sparse Discriminant Analysis. Therefore, in this paper, we add L_2,1-norm penalty term on FLDA and propose a feasible solution by transforming its nonlinear model into linear regression type. In addition, we modify the optimization model of SDA by replacing elastic net with L_2,1-norm penalty term and present its optimization method. Experiments on three standard face databases illustrate FLDA and SDA via L_2,1-norm penalty term can significantly improve their recognition performance, and obtain inspiring results with low computation cost and for low-dimension feature.     

COMPARING FACE IMAGES USING THE MODIFIED HAUSDORFF DISTANCE

We introduce a novel methodology applicable to face matching and fast screening of large facial databases. The proposed shape comparison method operates on edge maps and derives holistic similarity measures, yet, it does not require solving the point correspondence problem. While the use of edge images is important to introduce robustness to changes in illumination, the lack of point-to-point matching delivers speed and tolerance to local non-rigid distortions. In particular, we propose a face similarity measure derived as a variant of the Hausdorff distance by introducing the notion of a neighborhood function (N) and associated penalties (P). Experimental results on a large set of face images demonstrate that our approach produces excellent recognition results even when less than 3% of the original grey-scale face image information is stored in the face database (gallery). These results implicate that the process of face recognition may start at a much earlier stage of visual processing than it was earlier suggested. We argue, that edge-like retinal images of faces are initially screened “at a glance” without the involvement of high-level cognitive functions thus delivering high speed and reducing computational complexity.     

Sparse two-dimensional local discriminant projections for feature extraction

Two-dimensional local graph embedding discriminant analysis (2DLGEDA) and two-dimensional discriminant locality preserving projections (2DDLPP) were recently proposed to directly extract features form 2D face matrices to improve the performance of two-dimensional locality preserving projections (2DLPP). But all of them require a high computational cost and the learned transform matrices lack intuitive and semantic interpretations. In this paper, we propose a novel method called sparse two-dimensional locality discriminant projections (S2DLDP), which is a sparse extension of graph-based image feature extraction method. S2DLDP combines the spectral analysis and L₁-norm regression using the Elastic Net to learn the sparse projections. Differing from the existing 2D methods such as 2DLPP, 2DDLP and 2DLGEDA, S2DLDP can learn the sparse 2D face profile subspaces (also called sparsefaces), which give an intuitive, semantic and interpretable feature subspace for face representation. We point out that using S2DLDP for face feature extraction is, in essence, to project the 2D face images on the semantic face profile subspaces, on which face recognition is also performed. Experiments on Yale, ORL and AR face databases show the efficiency and effectiveness of S2DLDP.     

Face recognition via local preserving average neighborhood margin maximization and extreme learning machine

Average neighborhood maximum margin (ANMM) is an effective method for feature extraction in appearance-based face recognition. In this paper, we extend ANMM to locality preserving average neighborhood margin maximization (LPANMM) in order to maintain the local structure on the original data manifold in the discriminant feature space. We also combine LPANMM with extreme learning machine (ELM) as a new scheme for face recognition, we train the single-hidden layer feedforward neural network (SLFN) in the ELM classifier with the discriminant features that are extracted by LPANMM, then we use the trained ELM classifer to classify the test data. In the process of training SLFN, ELM can not only achieve the smallest training error in theory, but is also not sensitive to the initial value selection of the parameters for the SLFN. Experimental results on ORL, Yale, CMU PIE and FERET face databases demonstrate the scheme LPANMM/ELM can achieve better performance than ANMM and other traditional schemes for face recognition.     

Domain Adaptation for Face Recognition: Targetize Source Domain Bridged by Common Subspace

In many applications, a face recognition model learned on a source domain but applied to a novel target domain degenerates even significantly due to the mismatch between the two domains. Aiming at learning a better face recognition model for the target domain, this paper proposes a simple but effective domain adaptation approach that transfers the supervision knowledge from a labeled source domain to the unlabeled target domain. Our basic idea is to convert the source domain images to target domain (termed as targetize the source domain hereinafter), and at the same time keep its supervision information. For this purpose, each source domain image is simply represented as a linear combination of sparse target domain neighbors in the image space, with the combination coefficients however learnt in a common subspace. The principle behind this strategy is that, the common knowledge is only favorable for accurate cross-domain reconstruction, but for the classification in the target domain, the specific knowledge of the target domain is also essential and thus should be mostly preserved (through targetization in the image space in this work). To discover the common knowledge, specifically, a common subspace is learnt, in which the structures of both domains are preserved and meanwhile the disparity of source and target domains is reduced. The proposed method is extensively evaluated under three face recognition scenarios, i.e., domain adaptation across view angle, domain adaptation across ethnicity and domain adaptation across imaging condition. The experimental results illustrate the superiority of our method over those competitive ones.     

基于改进的多流形算法的人脸图像集识别

一个图像集由大量变化不一的图像组成,而且这些图像都表示同一个人.现实中的图像集数据是非线性的,造成这些现象的因素有人脸的角度不同、光线的明暗等,因此图像集中的每幅图像都是变化的,如果近似的将一个图像集建模为线性子空间,而忽略了集合中数据结构的变化,很显然是不合理的,这也必然会影响到最后的识别率.受流形理论知识的启发,可以将图像集建模为一个流形,这与传统的将图像集建模为子空间的方法有着本质区别.本文在基于流形的人脸图像集识别方法的基础上进行改进,提出新的计算样子空间距离方法,最后采用所有最短子空间距离的平均值作为流形之间的距离,称为改进的多流形方法(Improved multi-manifold method,IMM).IMM方法在CMU PIE数据库上进行实验,结果表明该方法相比其他方法具有更高识别率.