Hasan Serhan Yavuz

Large margin classifiers based on affine hulls

This paper introduces a geometrically inspired large margin classifier that can be a better alternative to the support vector machines (SVMs) for the classification problems with limited number of training samples. In contrast to the SVM classifier, we approximate classes with affine hulls of their class samples rather than convex hulls. For any pair of classes approximated with affine hulls, we introduce two solutions to find the best separating hyperplane between them. In the first proposed formulation, we compute the closest points on the affine hulls of classes and connect these two points with a line segment. The optimal separating hyperplane between the two classes is chosen to be the hyperplane that is orthogonal to the line segment and bisects the line. The second formulation is derived by modifying the @n-SVM formulation. Both formulations are extended to the nonlinear case by using the kernel trick. Based on our findings, we also develop a geometric interpretation of the least squares SVM classifier and show that it is a special case of the proposed method. Multi-class classification problems are dealt with constructing and combining several binary classifiers as in SVM. The experiments on several databases show that the proposed methods work as good as the SVM classifier if not any better.

Two-dimensional subspace classifiers for face recognition

The subspace classifiers are pattern classification methods where linear subspaces are used to represent classes. In order to use the classical subspace classifiers for face recognition tasks, two-dimensional (2D) image matrices must be transformed into one-dimensional (1D) vectors. In this paper, we propose new methods to apply the conventional subspace classifier methods directly to the image matrices. The proposed methods yield easier evaluation of correlation and covariance matrices, which in turn speeds up the training and testing phases of the classification process. Utilizing 2D image matrices also enables us to apply 2D versions of some subspace classifiers to the face recognition tasks, in which the corresponding classical subspace classifiers cannot be used due to high dimensionality. Moreover, the proposed methods are also generalized such that they can be used with the higher order image tensors. We tested the proposed 2D methods on three different face databases. Experimental results show that the performances of the proposed 2D methods are typically better than the performances of classical subspace classifiers in terms of recognition accuracy and real-time efficiency.

Automatic face recognition from frontal images

Face recognition can be described as identification of people from their face images. In this study, an automatic face recognition system has been designed by using frontal images photographed in our lab. The automatic face recognition procedure consists of an alignment process which includes face detection, eye detection, mapping of the center coordinates of the eyes to a standard face template. This is followed by classification of aligned faces. In literature, face alignment process is usually done with manually and high recognition rates can be achieved due to very well aligned faces. However, in real-time face recognition applications, its not possible to align face images manually. Therefore, successful classification rates reported in the literature are mostly misleading. In this study, we aligned faces in a fully automatic manner and we obtained more reliable and realistic face recognition rates. Face images are represented with gray level, LBP, LTP, and two dimensional Gabor filter features and performances are tested with Eigenfaces, Fisherfaces, and DCV methods. Experimental results showed that the automatic recognition rates can reach close to 90% correct recognition rates.

Comparisons of features for automatic eye and mouth localization

Localization of the eyes and mouth in face images is very important for accurate classification in automatic face recognition systems. The alignment of unknown face images with templates generally improves the performance of the face recognition system, and this process uses locations of the eyes and mouth. In this work, we compare different features (gray-level values, distance transform features, gradients and local binary patterns) for automatic localization of eyes and mouth. To this end, we use the sliding window approach using the linear and nonlinear support vector machine (SVM) classifiers. We created new frontal face data sets to train and test our algorithms. The experimental results show that the SVM classifier using the Gaussian kernel yields better results than the linear kernel. Among the four feature extraction methods, the performance of the local binary pattern features draws the attention for having better detection rates in both the linear and the nonlinear cases with smaller feature size.

A new distance measure for hierarchical clustering

Support vector machine (SVM) classifier formulation is originally designed for binary classification, and the extension of it to the multi-class case is still an open research problem. Classical approaches such as one-against-one or one-against-all have been used to address the multi-class problem, but these approaches become less appealing when the number of classes in the training set is too large. Recent approaches use hierarchical based classification for the multi-class problems since they scale well with the number of classes. SVM based hierarchical classifiers involve the partition of data samples through a clustering algorithm, and classification performance of the overall system heavily depends on the generated clusters. The clustering methods such as k-means, kernel k-means, spherical shells and balanced subset clustering have been used for this goal, but their distance measures, which are used for partitioning the data samples, are not compatible with the SVM classification goal. This paper introduces a new distance measure for partition of data samples for SVM based hierarchical classification. Unlike other clustering methods used for this goal, our proposed method is suitable when SVMs are used as the base classifier. As demonstrated in the experiments, integrating the proposed clustering scheme into the hierarchical SVM classifiers significantly improves the computational efficiency with a small decrease in the recognition accuracy.

Facial expression recognition based on locational features

Facial expression recognition is an active problem of behavioral science since Darwins work which was presented in 1872. Nowadays, development of imaging techniques and computer technology leads this problem to have various application areas like education of autistic children, non-verbal communication, human-machine interaction. In this study, we propose a novel feature extraction method to recognize seven facial expressions. After the detection of facial landmarks by using the supervised descent method, high level features are extracted from the variations of landmarks through the video frames. Recognition experiments have been performed on one of the most widely used datasets, namely CK+ database by using support vector machines in classification. Experimental results demonstrate that facial expressions have been successfully recognized.

Large Margin Classifier Based on Affine Hulls

This paper introduces a geometrically inspired large-margin classifier that can be a better alternative to the Support Vector Machines (SVMs) for the classification problems with limited number of training samples. In contrast to the SVM classifier, we approximate classes with affine hulls of their class samples rather than convex hulls, which may be unrealistically tight in high-dimensional spaces. To find the best separating hyperplane between any pair of classes approximated with the affine hulls, we first compute the closest points on the affine hulls and connect these two points with a line segment. The optimal separating hyperplane is chosen to be the hyperplane that is orthogonal to the line segment and bisects the line. To allow soft margin solutions, we first reduce affine hulls in order to alleviate the effects of outliers and then search for the best separating hyperplane between these reduced models. Multi-class classification problems are dealt with constructing and combining several binary classifiers as in SVM. The experiments on several databases show that the proposed method compares favorably with the SVM classifier.

Comparison of matrix decomposition and SIFT descriptor based methods for face alignment

Face alignment is an important pre-processing step for face analysis systems. Especially, the performance of face recognition systems can be improved by using aligned face images. In this work, we used matrix decomposition based and SIFT features based methods in face alignment. We performed recognition experiments by using raw versus aligned images with an image set based classification method. We also developed a SIFT-flow based affine transformation and showed that this type of alignment improves the recognition accuracies.

Towards Large-Scale Face Recognition Based on Videos

This paper introduces a new method to find the most important samples for classification in image sets to speed-up the classification phase and reduce the storage space for large-scale face recognition tasks that use image sets obtained from face videos. We approximate the image sets with the kernelized convex hulls and show that it is sufficient to use only the samples that participate to shape the image set boundaries in this setting. To find those important samples that form the image set boundaries in the feature space, we employed the kernelized Support Vector Data Description (SVDD) method which finds a compact hypersphere that fits the image set samples best. Then, we show that these kernelized hypersphere models can also be used to model image sets for classification purposes. Lastly, we introduce ESOGU-285 (ESkisehir OsmanGazi University) Face Videos database that includes 285 people since the most popular video datasets used for set based recognition methods include either a few amount of people or large amount of people with just a few (or single) video collections. The experimental results on small sized standard datasets and our new larger sized dataset show that the proposed method greatly improves the testing times of the classification system (we obtained speed-ups up to a factor of 10 in ESOGU Face Videos dataset) without a significant drop in accuracies.

One-sided best fitting hyperplane classifier

In this paper we propose a new hyperplane fitting classification method that does not have limitations of the existing hyperplane fitting classifiers. There are two principal improvements of the proposed method: It returns sparse solutions and it is suitable for large-scale problems. Both advantages are accomplished by using a simple trick, which constraints positive samples to lie between two parallel hyperplanes rather than to lie on a single fitting hyperplane. The experiments on several databases show that our proposed method typically outperforms other hyperplane fitting classifiers in terms of classification accuracy, and it produces comparable results to the Support Vector Machine classifier.