Shiladitya Chowdhury

Face recognition by generalized two-dimensional FLD method and multi-class support vector machines

This paper presents a novel scheme for feature extraction, namely, the generalized two-dimensional Fishers linear discriminant (G-2DFLD) method and its use for face recognition using multi-class support vector machines as classifier. The G-2DFLD method is an extension of the 2DFLD method for feature extraction. Like 2DFLD method, G-2DFLD method is also based on the original 2D image matrix. However, unlike 2DFLD method, which maximizes class separability either from row or column direction, the G-2DFLD method maximizes class separability from both the row and column directions simultaneously. To realize this, two alternative Fishers criteria have been defined corresponding to row and column-wise projection directions. Unlike 2DFLD method, the principal components extracted from an image matrix in G-2DFLD method are scalars; yielding much smaller image feature matrix. The proposed G-2DFLD method was evaluated on two popular face recognition databases, the AT&T (formerly ORL) and the UMIST face databases. The experimental results using different experimental strategies show that the new G-2DFLD scheme outperforms the PCA, 2DPCA, FLD and 2DFLD schemes, not only in terms of computation times, but also for the task of face recognition using multi-class support vector machines (SVM) as classifier. The proposed method also outperforms some of the neural networks and other SVM-based methods for face recognition reported in the literature.

Face Recognition by Fusing Local and Global Discriminant Features

This paper presents a novel scheme for face recognition by fusing local and global discriminant features. It has been observed that facial changes are occurred due to variations in facial expression, illumination condition, pose, etc. and these changes are often appeared only some regions of the whole image. The global features extracted from the whole image are not able to cope with these facial changes. To cope with the above facial changes face images are divided into a number of non-overlapping smaller sub-images and discriminant features are extracted from these sub-images as well as from the whole image. All these extracted local and global features are fused to form a large feature vector. We have used generalized two-dimensional fishers linear discriminate (G-2DFLD) method to extract these local and global discriminant features. We have used the fishers linear discriminate (FLD) method to extract lower dimensional discriminant features from the fused large feature vector. A Multi-class Support Vector Machine (SVM) is applied on these reduced feature vector for classification. The proposed method was evaluated on AT&T Face Database and experimental results show that the performance of the proposed method is better than other global feature extraction methods like PCA, 2DPCA, PCA+FLD, 2DFLD and G-2DFLD methods.

Feature extraction by fusing local and global discriminant features: An application to face recognition

This paper presents a novel scheme for feature extraction for face recognition by fusing local and global discriminant features. The facial changes due to variations of pose, illumination, expression, etc. are often appeared only some regions of the whole face image. Therefore, global features extracted from the whole image fail to cope with these variations. To address these problems, face images are divided into a number of non-overlapping sub-images and then G-2DFLD method is applied to each of these sub-images as well as to the whole image to extract local and global discriminant features, respectively. The G-2DFLD method is found to be superior to other appearance-based methods for feature extraction. All these extracted local and global discriminant features are then fused to get a large feature vector. Its dimensionality is then reduced by the PCA technique to decrease overall complexity of the system. A multi-class SVM is used as a classifier for recognition based on these reduced features. The proposed method was evaluated on two popular face recognition databases, the AT&T (formerly ORL) and the UMIST face databases. The experimental results show that the new method outperforms the global features extracted by the PCA, 2DPCA, PCA+FLD, 2DFLD and G-2DFLD methods in terms of face recognition.

An improved hybrid approach to face recognition by fusing local and global discriminant features

This paper presents a novel scheme for face recognition by fusing local and global discriminant features. It has been observed that facial changes are occurred due to variations in facial expression, illumination condition, pose, etc. and these changes are often appeared only some regions of the whole image. The global features extracted from the whole image are not able to cope with these facial changes. To cope with the above facial changes face images are divided into a number of non-overlapping smaller sub-images and discriminant features are extracted from these sub-images as well as from the whole image. All these extracted local and global features are fused to form a large feature vector. We have used generalized two-dimensional fishers linear discriminate (G-2DFLD) method to extract these local and global discriminant features. We have used the fishers linear discriminate (FLD) method to extract lower dimensional discriminant features from the fused large feature vector. A Multi-class Support Vector Machine (SVM) is applied on these reduced feature vector for classification. The proposed method was evaluated on AT&T Face Database and experimental results show that the performance of the proposed method is better than other global feature extraction methods like PCA, 2DPCA, PCA+FLD, 2DFLD and G-2DFLD methods.

Generalized two-dimensional FLD method for feature extraction: an application to face recognition

This paper presents a novel scheme for face feature extraction, namely, the generalized two-dimensional Fishers linear discriminant (G-2DFLD) method The G-2DFLD method is an extension of the 2DFLD method for feature extraction Like 2DFLD method, G-2DFLD method is also based on the original 2D image matrix However, unlike 2DFLD method, which maximizes class separability either from row or column direction, the G-2DFLD method maximizes class separability from both the row and column directions simultaneously In G-2DFLD method, two alternative Fishers criteria have been defined corresponding to row and column-wise projection directions The principal components extracted from an image matrix in 2DFLD method are vectors; whereas, in G-2DFLD method these are scalars Therefore, the size of the resultant image feature matrix is much smaller using G-2DFLD method than that of using 2DFLD method The proposed G-2DFLD method was evaluated on two popular face recognition databases, the AT&T (formerly ORL) and the UMIST face databases The experimental results show that the new G-2DFLD scheme outperforms the PCA, 2DPCA, FLD and 2DFLD schemes, not only in terms of computation times, but also for the task of face recognition using a multi-class support vector machine.

A Novel Elastic Window for Face Detection and Recognition from Video

Since the past several years, face recognition from video has received significant attention due to wide range of commercial and law enforcement applications, such as surveillance systems, closed circuit TV (CCTV) monitoring, etc. Human face detection is the first and important task in a dynamic environment, such as video, where noise conditions, illuminations, locations of subjects and pose can vary significantly from one frame to another frame. In this paper, a novel elastic window, which does not make any assumption about the pose, expression or prior localization of a face in a video frame is presented for finding boundary of face region. The window locates the possible face boundaries by elastically expanding its size using local image gradients. Prior to this, a video-frame undergoes in several pre-processing tasks in order to remove noise, background, etc. And producing thin binary image representing only possible face boundaries and scattered noises. After detecting faces from video frames, we extract discriminant facial features from these cropped face images. A multi-class SVM is used as a classifier for face recognition based on these facial features. The proposed method was evaluated on Honda/UCSD video database and the experimental results show that the proposed method outperforms several existing video-based face recognition methods in terms of face recognition.

An efficient method of face recognition by fusing original and diagonal images

Image level fusion combines an image in different ways with its original version so that the combine image may contain more relevant information than the original one. This paper presents a novel method for face recognition by fusing original and corresponding diagonal images. Two ways of image fusion technique have been performed here. Firstly, we generate diagonal face image from original face image and append original image with the diagonal image horizontally side by side. Secondly, original face image and the corresponding diagonal face image are appended vertically to get two sets of fused image matrices. The G-2DFLD method is applied on both of these large fused images for extraction of discriminant features, which integrate the underlying discriminant information along the horizontal, vertical and diagonal directions. This extracted feature matrices are applied on Radial Basis Function-Neural Networks (RBF-NN) for classification and recognition. Experiments on the AT&T face database (formally known as ORL database) indicate the superiority of the proposed method as compared to some of the conventional methods.

Weighted Multi-Class Support Vector Machine for robust face recognition

This paper presents a novel scheme for face recognition using Weighted Multi-class Support Vector Machine (WMSVM). Support Vector Machine (SVM) is well-known powerful tool for solving classification problem. Weighted Support Vector Machines (Weighted SVM) are extension of the SVM. It has been seen that different input vectors make different contribution to the learning of a decision surface. Therefore, different weights are assigned to different data points, so that the Weighted SVM training algorithm learns the decision surface according to the relative importance of data points in the training data. In our proposed WMSVM, probabilistic method is used for weight generation. The generalized two-dimensional Fishers linear discriminant (G-2DFLD)-based facial features are applied on the proposed WMSVM for recognition. The experimental results on UMIST and AR face database show that the proposed Weighted Multi-class SVM yields higher recognition rate than standard Multi-class SVM.

Feature Extraction Using Fuzzy Generalized Two-Dimensional Inverse LDA with Gaussian Probabilistic Distribution and Face Recognition

This paper proposes a feature extraction technique called Gaussian probabilistic fuzzy generalized two-dimensional Fisher’s inverse linear discriminant analysis (GPFG-2DILDA) method based on fuzzy set theory, Gaussian probabilistic distribution information and inverse LDA. Like the FG-2DLDA, the proposed GPFG-2DILDA method also maximizes class separability along x- and y-axes directions simultaneously. The proposed method first calculates fuzzy membership matrix by fuzzy k-nearest neighbour (Fk-NN) algorithm. These values are combined with the training samples to obtain the global mean and class-wise mean training images. Thereafter, fuzzy membership values are integrated into intra-class and inter-class scatter matrices along both (x- and y-) directions. Similarly, Gaussian probabilistic distribution information is incorporated into the intra-class scatter matrices. Finally, by solving the eigenvalue problems of these scatter matrices, we find the optimal Gaussian-fuzzy inverse projection vectors, which actually used to generate more discriminant features and to solve the binary classification problem. The GPFG-2DILDA method has been evaluated on the AT&T face database to demonstrate the efficacy of the proposed method over some state-of-the-art face recognition methods.

An Efficient Face Recognition Method by Fusing Spatial Discriminant Facial Features

Feature level fusion is a very well known technique for improving the performance of a face recognition system. This paper presents an approach of fusion of directional spatial discriminant features for face recognition. The key idea of the proposed method is to fuse the facial features lying along the horizontal, vertical and diagonal directions, so that this fused feature vector can contain more discriminant information than the individual facial feature of single direction only. However due to the fusion of features the size of fused feature vector becomes larger, which may increase complexity of the classifier to be used for recognition. To optimize this lower dimensional discriminant features are again extracted from this large fused feature vector. In our experiment we have applied G-2DFLD method on the original images to extract the discriminant features. Then original images are converted into diagonal images and another set of discriminant features, representing the diagonal information, are extracted by using the G-2DFLD method. The original and diagonal feature matrices are then fused to form a large feature matrix. The dimension of this large fused matrix is then further reduced by G-2DFLD method and this resultant matrix is used for classification and recognition by Radial Basis Function-Neural Networks (RBF-NN). Experiments on the AT&T (formally known as ORL database) face database indicate the competitive performance of the proposed method, as compared to some existing subspaces-based methods.