Yale Face Database Research Articles

HIERARCHICAL STRUCTURE BASED CONVOLUTIONAL NEURAL NETWORK FOR FACE RECOGNITION

In this paper, a hierarchical structure based convolutional neural network is proposed to provide the ability for robust information processing. The weight sharing ability of convolutional neural networks (CNNs) is considered as a level of hierarchy in these networks. Weight sharing reduces the number of free parameters and improves the generalization ability. In the proposed structure, a small CNN which is used for feature extractor is shared between the whole input image pixels. A scalable architecture for implementing extensive CNNs is resulted using a smaller and modularized trainable network to solve a large and complicated task. The proposed structure causes less training time, fewer numbers of parameters and higher test data accuracy. The recognition accuracy for recognizing unseen data shows improvement in generalization. Also presented are application examples for face recognition. The comprehensive experiments completed on ORL, Yale and JAFFE face databases show improved classification rates and reduced training time and network parameters.

Locality Preserving Maximum Scatter Difference Projection for Face Recognition

Maximum Scatter Difference (MSD) aims to preserve discriminant information of sample space, but it fails to find the essential structure of the samples with nonlinear distribution. To overcome this problem, an efficient feature extraction method named as Locality Preserving Maximum Scatter Difference (LPMSD) projection is proposed in this paper. The new algorithm is developed based on locality preserved embedding and MSD criterion. Thus, the proposed LPMSD not only preserves discriminant information of sample space but also captures the intrinsic submanifold of sample space. Experimental results on ORL, Yale and CAS-PEAL face database indicate that the LPMSD method outperforms the MSD, MMSD and LDA methods under various experimental conditions.

Median–mean line based discriminant analysis

This paper presents a median–mean line based discriminant analysis (MMLDA) technique for dimensionality reduction. Taking the negative effect on the class-mean caused by outliers into account, MMLDA introduces the median–mean line (MML) as an adaptive class-prototype. Based on the MML, the point-to-MML distance is designed and used as the measure metric to characterize the within-class median–mean linear scatter as well as the between-class median–mean linear scatter. Such a characterization makes MMLDA more robust than many class-mean based methods, like classical Fisher linear discriminant analysis (FLDA). In addition, the connection between MMLDA and FLDA is presented in this paper. Finally, the proposed method is evaluated using the AR face database, the Yale face database, the UCI Wine database and the ETH80 object category database. The experimental results demonstrate the effectiveness of MMLDA.

Semi-supervised Discriminant Analysis Based on Maximum Scatter Difference

For the problems of use of a few marked samples and small sample size problem in face recognition, a new algorithm of semi-supervised discriminant analysis based on maximum scatter difference is proposed. Based on MFA algorithm, the algorithm was improved semi-supervised by adding UDP local and non local scatter matrix to the target function. This could make use of the not marked samples that supervised algorithm wasted and the classification label information which unsupervised algorithm does not use. At the same time this solves to the problem that can not direct matrix inversion for scatter matrix singular by introducing the maximum scatter difference. Finally, the effectiveness of the proposed methods is validated through the experimental results on ORL and YALE face databases.

A novel face feature descriptor using adaptively weighted extended LBP pyramid

This paper proposes a new face description with single sample by Adaptively Weighted Extended Local Binary Pattern Pyramid (AWELBPP). First, the proposed algorithm utilizes pyramid transform to represent sample image into multi-resolution images. Second, the multi-resolution images are divided into a set of horizontal sub-images. Then, Extended Local Binary Pattern (ELBP) is applied to the sub-images in order to calculate the Sub-ELBP pyramid and the local image entropy is employed to the sub-images for an adaptively weighting map (AWM) that can measure the importance of the information they contain. Finally, AWELBPP feature is extracted from the fusion of the Sub-ELBP pyramid and the AWM. Under different illumination, facial expression and partial occlusion conditions, simulated experiments and comparisons on many subsets of Yale face databases, Yale B face databases and ORL face databases show that the proposed algorithm is an outstanding method for single sample face recognition compared with local PCA, ELBP, CLBP, PLBP.

A Fuzzy Adaptive K-SVD Dictionary Algorithm for Face Recogntion

Sparse representations using overcomplete dictionaries has concentrated mainly on the study of pursuit algorithms that decompose signals with respect to a given dictionary. Designing dictionaries to better fit the above model can be done by either selecting one from a prespecified set of linear transforms or adapting the dictionary to a set of training signals. The K-SVD algorithm is an iterative method that alternates between sparse coding of the examples based on the current dictionary and a process of updating the dictionary atoms to better fit the data. However, the existing K-SVD algorithm is employed to dwell on the concept of a binary class assignment meaning that the multi-classes samples are assigned to the given classes definitely. The work proposed in this paper provides a novel fuzzy adaptive way to adapting dictionaries in order to achieve the fuzzy sparse signal representations, the update of the dictionary columns is combined with an update of the sparse representations by incorporated a new mechanism of fuzzy set, which is called fuzzy K-SVD. Experimental results conducted on the ORL and Yale face databases demonstrate the effectiveness of the proposed method.

A New Construction Method of Neighbor Graph Based on Correlative Columns Information for Marginal Fisher Analysis

Marginal fisher analysis is a typical supervised method which has been used in many practical problems such as face recognition. However, MFA mainly depends on its essential neighbor graphsintrinsic graph and penalty graph. Intrinsic graph characterizes the intra-class compactness while the inter-class graph characterizes the inter-class separability. Consequently, neighbor graph construction plays a vital role on the performance of MFA. In this paper, we propose a new construction method of intrinsic graph and penalty graph for marginal fisher analysis. It is based on correlative columns information, so we name this new method as Correlative Columns Information based MFA (CCIMFA). CCIMFA can well show the spatial structure information of the original image matrices, and also can preserve the corresponding columns information. CCIMFA also has anther attractive property that is columns’ noise immunity. In order to test and evaluate CCIMFA’s performance, a series of experiments were performed on the well-known face databases: ORL and Yale face databases. The experimental results show that CCIMFA achieves better performance than MFA.

Illumination normalization using fuzzy filter in DCT domain for face recognition

We develop a new approach of illumination normalization for face recognition under varying lighting conditions. The effect of illumination variations is in decreasing order over low-frequency discrete cosine transform (DCT) coefficients. The proposed approach is expected to nullify the effect of illumination variations as well as to preserve the low-frequency details of a face image in order to achieve a good recognition performance. This has been accomplished by using a fuzzy filter applied over the low-frequency DCT (LFDCT) coefficients. The ‘simple classification technique’ (k-nearest neighbor classification) is used to establish the performance improvement by present approach of illumination normalization under high and unpredictable illumination variations. Our fuzzy filter based illumination normalization approach achieves zero error rate on Yale face database B (named as Yale B database in this work) and CMU PIE database. An excellent performance is achieved on extended Yale B database. The present approach of illumination normalization is also tested on Yale face database which comprises of illumination variations together with expression variations and misalignment. Significant reduction in the error rate is achieved by the present approach on this database as well. These results establish the superiority of the proposed approach of illumination normalization, over the existing ones.

A parameterized fuzzy adaptive K-SVD approach for the multi-classes study of pursuit algorithms

Sparse representations using over complete dictionaries has concentrated mainly on the study of pursuit algorithms that decompose signals with respect to a given dictionary. Designing dictionaries to better fit the above model can be done by either selecting one from a pre-specified set of linear transforms or adapting the dictionary to a set of training signals. The K-SVD algorithm is an iterative method that alternates between sparse coding of the examples based on the current dictionary and a process of updating the dictionary atoms to better fit the data. However, the existing K-SVD algorithm is employed to dwell on the concept of a binary class assignment, which means that the multi-classes samples are assigned to the given classes definitely. The work proposed in this paper provides a novel parameterized fuzzy adaptive way to adapting dictionaries. In order to achieve the fuzzy sparse signal representations, the update of the dictionary columns is combined with an update of the sparse representations by embedding a new mechanism of fuzzy set, which is called parameterized fuzzy K-SVD. Experimental results conducted on the ORL, Yale and FERET face databases demonstrate the effectiveness of the proposed method.

A fast classification scheme and its application to face recognition

To overcome the high computational complexity in real-time classifier design, we propose a fast classification scheme. A new measure called ‘reconstruction proportion’ is exploited to reflect the discriminant information. A novel space called the ‘reconstruction space’ is constructed according to the reconstruction proportions. A point in the reconstruction space denotes the case of a sample reconstructed using training samples. This is used to search for an optimal mapping from the conventional sample space to the reconstruction space. When the projection from the sample space to the reconstruction space is obtained, a new sample after mapping to the new discriminant space would be classified quickly according to the reconstruction proportions in the reconstruction space. This projection technique results in a diversion of time-consuming calculations from the classification stage to the training stage. Though training time is prolonged, it is advantageous in that classification problems such as identification can be solved in real time. Experimental results on the ORL, Yale, YaleB, and CMU PIE face databases showed that the proposed fast classification scheme greatly outperforms conventional classifiers in classification accuracy and efficiency.

Fuzzy MSD based feature extraction method for face recognition

To improve the recognition performance of maximum scatter difference (MSD), a fuzzy MSD method is proposed in this paper. In the existing MSD model, the class mean vector in the expressions of within-class scatter matrix and between-class scatter matrix is estimated by class sample average. Obviously, the class sample average is not sufficient to provide an accurate estimate of the class mean using a few of the given samples, because there will be some outliers in the sample set under the non-ideal conditions such as variations of expression, illumination, pose, and so on. As a result, the recognition performance of traditional MSD model will decrease. To address this problem, inspired by existing fuzzy application, the fuzzy theory is incorporated into traditional maximum scatter difference algorithm. In this method, applying fuzzy K-nearest neighbor (FKNN), the membership degree matrix of training samples is calculated, which is used to get fuzzy means of each class and the average of fuzzy means is then applied to the definition of within-class scatter matrix and between class scatter difference matrix, respectively. The results of experiments conducted on ORL, YALE and FERET face database indicate the effectiveness of the proposed approach.

Orthogonal discriminant improved local tangent space alignment based feature fusion for face recognition

Improved local tangent space alignment (ILTSA) is a recent nonlinear dimensionality reduction method which can efficiently recover the geometrical structure of sparse or non-uniformly distributed data manifold. In this paper, based on combination of modified maximum margin criterion and ILTSA, a novel feature extraction method named orthogonal discriminant improved local tangent space alignment (ODILTSA) is proposed. ODILTSA can preserve local geometry structure and maximize the margin between different classes simultaneously. Based on ODILTSA, a novel face recognition method which combines augmented complex wavelet features and original image features is developed. Experimental results on Yale, AR and PIE face databases demonstrate the effectiveness of ODILTSA and the feature fusion method.

An Embedded Feature Extraction Algorithm with MMC for Face Recognition

Feature extraction is the most crucial part of face recognition system, which has important role in the field of pattern recognition. There are a lot of classic algorithms about feature extraction at present, such as the method based on linear and nonlinear. An effective classification and dimension reduction method, local embedded graph attribute selection algorithm with maximal region is to generate the inherent graph and penalty graph, to overcome the low efficiency problem of local linear embedding (LLE) method and maximum margin criterion (MMC) method. With inherent picture, the structured nonlinear can be found on the high-dimensioned space by using local geometry of the restructured linear, which leads to the same instances gathering together as more as possible. Meanwhile, different class instances are as far as possible from each other in penalty picture. Since LLE is an unsupervised method, not enhance visual clustering classification ability, so compact figure within the class to consider the sample class information, can sample the same category as compact. In this method, the smallest size instance issue was tackled by the employment of MMC and the neighborhood relationship can be better described by an adequate improvement of the adjacency matrix. The effectiveness of algorithm proposed by this paper is present by a large amount of experiences in the standard face databases of Yale, AR and ORL facial data.

Sparse tensor discriminant analysis

The classical linear discriminant analysis has undergone great development and has recently been extended to different cases. In this paper, a novel discriminant subspace learning method called sparse tensor discriminant analysis (STDA) is proposed, which further extends the recently presented multilinear discriminant analysis to a sparse case. Through introducing the L1 and L2 norms into the objective function of STDA, we can obtain multiple interrelated sparse discriminant subspaces for feature extraction. As there are no closed-form solutions, k-mode optimization technique and the L1 norm sparse regression are combined to iteratively learn the optimal sparse discriminant subspace along different modes of the tensors. Moreover, each non-zero element in each subspace is selected from the most important variables/factors, and thus STDA has the potential to perform better than other discriminant subspace methods. Extensive experiments on face databases (Yale, FERET, and CMU PIE face databases) and the Weizmann action database show that the proposed STDA algorithm demonstrates the most competitive performance against the compared tensor-based methods, particularly in small sample sizes.

Compressed sensing with MCT and I(2D)2PCA processing for efficient face recognition

This article describes an effective human face recognition algorithm. Even though the principle component analysis (PCA) is one of the most common feature extraction methods, it is not suitable to implement a real-time embedded system for face recognition because large amount of computational load and memory capacity are necessary. To overcome this problem, we employ the incremental two-directional two-dimensional PCA (I(2D)2PCA) which is a combination of the (2D)2PCA to demand much less computational complexity than the conventional PCA and the incremental PCA (IPCA) to adapt the eigenspace only by using a new incoming sample datum without reusing of all the previous trained data. Furthermore, the modified census transform (MCT), a local normalization method useful for real-world application and implementation in an embedded system, is adopted to address robustness to illumination variations. To achieve better recognition accuracy with less computational load, the processed features are classified by the compressive sensing approach using l2–minimization. Experimental results on the Yale Face Database B show that the described system using the l2–minimization-based classification method for input data processed by the I(2D)2PCA and the MCT provided efficient and robust face recognition. © 2013 Wiley Periodicals, Inc. Int J Imaging Syst Technol, 23, 133–139, 2013

Face recognition algorithm based on multi-level texture spectrum features and PCA

To improve the recognition rate of Principal Component Analysis(PCA) algorithm in face recognition,a new algorithm combining the image texture spectrum feature with PCA was proposed.Firstly,the texture unit operator was used to extract the texture spectrum feature of the face image.Secondly,PCA approach was used to reduce the dimensions of the texture spectrum feature.Finally,K-Nearest Neighbor(KNN) classification was chosen to recognize the face.ORL and Yale face database were used to test the proposed algorithm,and the recognition accuracies were 96.5% and 95% respectively,which were higher than those of PCA and Modular Two-Dimensional PCA(M2DPCA).The experimental results demonstrate the efficiency and accuracy of the proposed algorithm.

A New Construction Method of Neighbor Graph for Locality Preserving Projections

In this paper, a new construction method of neighbor graph is proposed for Locality Preserving Projections (LPP). LPP is a typical method of graph-based dimensionality reduction, and has been successfully applied in many practical problems such as face recognition. However, LPP mainly depends on its essential neighbor graph whose construction suffers from the following issues: the data set is vectorized to compute the k-nearest neighbor graph (adjacency graph), which leads to the lost of the correlative columns information. So, we propose a new neighbor graph construction method which can well show the spatial structure information of the original image matrices, to preserve the corresponding columns information. In order to test and evaluate our method’s performance, a series of experiments were performed on the well-known face databases: ORL and Yale face databases. The experimental results show that our method achieves better performance than LPP.

Optimal Feature Extraction Using Greedy Approach for Random Image Components and Subspace Approach in Face Recognition

An innovative and uniform framework based on a combination of Gabor wavelets with principal component analysis (PCA) and multiple discriminant analysis (MDA) is presented in this paper. In this framework, features are extracted from the optimal random image components using greedy approach. These feature vectors are then projected to subspaces for dimensionality reduction which is used for solving linear problems. The design of Gabor filters, PCA and MDA are crucial processes used for facial feature extraction. The FERET, ORL and YALE face databases are used to generate the results. Experiments show that optimal random image component selection (ORICS) plus MDA outperforms ORICS and subspace projection approach such as ORICS plus PCA. Our method achieves 96.25%, 99.44% and 100% recognition accuracy on the FERET, ORL and YALE databases for 30% training respectively. This is a considerably improved performance compared with other standard methodologies described in the literature.

Global and Local Information Based Spherical Marginal Fisher Analysis for Face Recognition

We proposed a new face recognition algorithm, termed Spherical Marginal Fisher Analysis (SMFA). Different from traditional Marginal Fisher Analysis (MFA) in which we don’t select a certain number of nearest samples between different classes, but contain all the needed samples in some content. Meanwhile, we add the information between sample centers as applied in Linear Discriminant Analysis (LDA). Experimental results on the ORL and Yale face databases show our method outperforms other linear methods.

Maximum margin sparse representation discriminative mapping with application to face recognition

Sparse subspace learning has drawn more and more attention recently. We propose a novel sparse subspace learning algorithm called maximum margin sparse representation discriminative mapping (MSRDM), which adds the discriminative information into sparse neighborhood preservation. Based on combination of maximum margin discriminant criterion and sparse representation, MSRDM can preserve both local geometry structure and classification information. MSRDM can avoid the small sample size problem in face recognition naturally and the computation is efficient. To improve face recognition performance, we propose to integrate Gabor-like complex wavelet and natural image features by complex vectors as input features of MSRDM. Experimental results on ORL, UMIST, Yale, and PIE face databases demonstrate the effectiveness of the proposed face recognition method.