Face verification has many potential applications including filtering and ranking image/video search results on celebrities. Since these images/videos are taken under uncontrolled environments, the problem is very challenging due to dramatic lighting and pose variations, low resolutions, compression artifacts, etc. In addition, the available number of training images for each celebrity may be limited, hence learning individual classifiers for each person may cause overfitting. In this paper, we propose two ideas to meet the above challenges. First, we propose to use individual bins, instead of whole histograms, of Local Binary Patterns (LBP) as features for learning, which yields significant performance improvements and computation reduction in our experiments. Second, we present a novel Multi-Task Learning (MTL) framework, called Boosted MTL, for face verification with limited training data. It jointly learns classifiers for multiple people by sharing a few boosting classifiers in order to avoid overfitting. The effectiveness of Boosted MTL and LBP bin features is verified with a large number of celebrity images/videos from the web. 1.

Linear dimensionality reduction (LDR) techniques are quite important in pattern recognition due to their linear time complexity and simplicity. In this paper, we present a novel LDR technique which, though linear, aims to maximize the Chernoff distance in the transformed space; thus, augmenting the class separability in such a space. We present the corresponding criterion, which is maximized via a gradient-based algorithm, and provide convergence and initialization proofs. We have performed a comprehensive performance analysis of our method combined with two well-known classifiers, linear and quadratic, on synthetic and real-life data, and compared it with other LDR techniques. The results on synthetic and standard real-life datasets show that the proposed criterion outperforms the latter when combined with both linear and quadratic classifiers.

We propose a learning framework that actively explores creation of face space(s) by selecting images that are complementary to the images already represented in the face space. We also construct ensembles of classifiers learned from such actively sampled image sets, which further provides improvement in the recognition rates. We not only significantly reduce the number of images required in the training set but also improve the accuracy over learning from all the images. We also show that the single face space or ensemble of face spaces, thus constructed, has a higher generalization performance across different illumination and expression conditions.

Abstract — Face recognition has been of interest to a growing number of researchers due to its applications on security. Within past years, numerous face recognition algorithms have been proposed by researchers. However, there is no evidence that shows one specific proposed method is the best under all circumstances. So, a combination of several methods can be a good approach. Committee machine structures, which were introduced in the machine learning community, show some ways to combine different methods in a single framework. A committee machine structure makes decision according to its components. In the previous face recognition methods with committee machines, only the information which is extracted from test phase is used for combining classifiers. In this paper, in addition to the information in test phase, training phase information is used for combining classifiers. For this purpose, we introduce a new unit which is called “Region Finder”. This unit is attached to each classifier in a committee machine structure and is learned based on train phase information. A region finder determines its classifier recognition power in the classifier feature space. We applied our idea to a structure of five well-known classifiers, PCA, ICA, LDA, SVM and neural networks which are implemented for face recognition. Comparative experimental results of our committee machine with different algorithms and the structure without region finder units, demonstrate that the proposed system achieves improved accuracy.

Abstract — In this paper, we propose a novel ensemble-based approach to boost performance of traditional face recognition methods. The ensemble-based approach is based on the recently emerged technique known as “boosting. ” However, it is generally believed that boosting-like learning rules are not suited to a strong and stable learner such as LDA or PCA. To break the limitation, a novel weakness analysis theory is developed in this paper. This theory attempts to increase the diversity between the classifiers train set. For discriminating classifiers in the structure, a train set is divided into some subsets according to dependency or independency of train classes. Then each classifier will be learned on each of these non-overlap train sets. We call a train set is independent (or dependent), if its member, which are face classes, are maximally unsimlar (similar). We use graphs for getting dependent or independent sets of face classes. For combining classifiers, a new unit, which is called “Region Finder”, is introduced. This unit indicates the power of a classifier in the classifier feature space. According to dependent or independent sets, two architectures are proposed which each of them has special characteristics. Promising experimental results obtained on various difficult face recognition scenarios demonstrate the effectiveness of the proposed approach. We believe that this work is especially beneficial in extending the boosting framework to accommodate general (strong/weak) learners.

Face recognition has been a very active research area for past two decades due to its widely applications such as identity authentication, airport security and access control, surveillance, and video retrieval systems, etc. Numerous approaches have been proposed for face recognition and considerable successes have been reported [1]. A successful face recognition