Ajay Jaiswal

A Hybrid of Principal Component Analysis and Partial Least Squares for Face Recognition across Pose

In this paper, we propose a simple and efficient hybrid approach based on the combination of principal component analysis and partial least squares. Principal component analysis is used to reduce the dimension of image in first step and partial least squares method is used to carry out pose invariant face recognition in second step. The performance of proposed method is compared with another popular method based on global linear regression on hybrid-eigenface (HGLR) in terms of classification accuracy and computation time. Experimental results on two well known publicly available face databases demonstrate the effectiveness of the proposed approach.

Illumination Invariant Facial Pose Classification

this paper, we compared the performance of various combinations of edge operators and linear subspace methods to determine the best combination for pose classification. To evaluate the performance, we have carried out experiments on CMU-PIE database which contains images with wide variation in illumination and pose. We found that the performance of pose classification depends on the choice of edge operator and linear subspace method. The best classification accuracy is obtained with Prewitt edge operator and Eigenfeature regularization method. In order to handle illumination variation, we used adaptive histogram equalization as a preprocessing step resulting into significant improvement in performance except for Roberts operator. General Terms Pattern Recognition, Classification. KeywordsClassification, Edge detection, Linear Subspace Methods.

Performance evaluation of linear subspace methods for face recognition under illumination variation

Due to high dimensionality of face images and finite number of training samples, the linear subspace technique for face recognition pose challenges for better performance. In this paper, we compare the performance of linear subspace methods which involve the computation of scatter matrices for face recognition under illumination variation. The performance of these methods is evaluated in terms of classification accuracy, computational training and testing time. Extensive empirical experiments are performed to compare the performance using AR, Pie, Yale and YaleB face databases. In absence of sufficient number of training samples, classification accuracy of linear subspace methods deteriorate. Experimental results show that the performance of Dual LDA is best in terms of average classification accuracy. It is also observed that Fisherface takes minimum training time and both ERE and SVM takes minimum testing time. No linear subspace method outperforms others in terms of all performance measures.

Salient Object Detection in Noisy Images

Salient Object Detection SOD has several applications including image and video compression, video summarization, image segmentation and object discovery etc. Several Methods have been suggested in literature for detecting salient object in digital images. Most of these methods aim at detecting salient objects in images which does not contain any artifact such as noise. In this paper, we have evaluated several salient object detection methods in noisy environment on publicly available ASD Dataset. The performance of the methods is evaluated in terms of Precision, Recall and F-measure and Area under the curve AUC. It has been observed that there is no clear winner but the methods proposed by Liu et al. and Harel et al. are better in comparison to other methods.

A Comparative Study of Linear Discriminant and Linear Regression Based Methods for Expression Invariant Face Recognition

In the literature, the performance of Fisher’s Linear Discriminant (FLD), Linear Regression (LR) and their variants is found to be satisfactory for face recognition under illumination variation. However, face recognition under expression variation is also a challenging problem and has received little attention. To determine suitable method for expression invariant face recognition, in this paper, we have investigated several methods which are variants of FLD or LR. Extensive experiments are performed on three publicly available datasets namely ORL, JAFFE and FEEDTUM with varying number of training images per person. The performance is evaluated in terms of average classification accuracy. Experimental results demonstrate superior performance of Enhanced FLD (EFLD) method in comparison to other methods on all the three datasets. Statistical ranking used for comparison of methods strengthen the empirical findings.

Local Linear Regression on Hybrid Eigenfaces for Pose Invariant Face Recognition

Pose variation leads to significant decline in the performance of the face recognition systems. In this paper, the authors propose a new approach HLLR, based on conjunction of hybrid-eigenfaces and local linear regression LLR, to perform face recognition across pose. In this approach, LLR on hybrid-eigenfaces is used to generate virtual views. These virtual views in frontal and non-frontal poses are obtained using frontal gallery image. The performance of the proposed approach is compared for classification accuracy with another efficient method based on global linear regression on hybrid eigenface HGLR. They also investigate the effect of number of images used to construct hybrid-eigenfaces on classification accuracy. Experimental results on two well known publicly available face databases demonstrate the effectiveness of the proposed approach. The suitability of proposed approach is also noticed when the number of available images is small.

Analysis and evaluation of regression-based methods for facial pose classification

Facial pose classification is one of the important steps in some pose invariant face recognition methods. Regression has been used for facial pose classification. In this paper, facial pose classification approaches using different types of regression are compared in terms of average classification accuracy and computation time. We also analyse the time complexity of regression-based approaches for pose classification. Performance of these approaches is also compared with other popular approaches in terms of classification accuracy. Experimental results on two publicly available face databases (PIE and FERET) show that the performance of regression-based approaches is comparable and generally outperform other approaches. Among regression-based methods, local linear regression with overlap outperforms other methods. In terms of computation time, global linear regression and nonlinear regression are comparable and better than others. We also analysed the performance of regression-based approaches after adding Gaussian noise with zero mean in test images and found that global linear regression and nonlinear regression-based approaches perform better than other.

Incremental and Decremental Exponential Discriminant Analysis for Face Recognition

Linear Discriminant Analysis (LDA) is widely used for feature extraction in face recognition but suffers from small sample size (SSS) problem in its original formulation. Exponential discriminant analysis (EDA) is one of the variants of LDA suggested recently to overcome this problem. For many real time systems, it may not be feasible to have all the data samples in advance before the actual model is developed. The new data samples may appear in chunks at different points of time. In this paper, the authors propose incremental formulation of EDA to avoid learning from scratch. The proposed incremental algorithm takes less computation time and memory. Experiments are performed on three publicly available face datasets. Experimental results demonstrate the effectiveness of the proposed incremental formulation in comparison to its batch formulation in terms of computation time and memory requirement. Also, the proposed incremental algorithms (IEDA, DEDA) outperform incremental formulation of LDA in terms of classification accuracy.

Statistical framework for facial pose classification

Pose classification is one of the important steps in some pose invariant face recognition methods. In this paper, we propose to use: (i) Partial least square (PLS) and (ii) Linear regression for facial pose classification. The performance of these two approaches is compared with two edge based approaches and pose-eigenspace approach in terms of classification accuracy. Experimental results on two publicly available face databases (PIE and FERET) show that the regression based approach outperforms other approaches for both the databases.