A novel training weighted ensemble (TWE) with application to face recognition
Introduction
Many researchers have solved pattern recognition problems by introducing solutions that are based on a single classifier. In their studies, that unique classifier has been trained very well, and then used to recognize the unseen (test) instances. For each test instance, the decision that is taken by this classifier is considered to be the final decision of the introduced solution. The design of a unique classifier that results in excellent percentage of correct classification is very complex especially when each instance has a huge number of attributes, which is the case in face recognition problems. In [16], the problems that face the design of that unique classifier are discussed in detail.
On the other hand, a lot of studies gave other solutions to pattern recognition problems which are based on a combination of multiple classifiers. In their researches, the final decision of the whole solution is taken after combining the decisions of all the individual classifiers. These studies agreed that the decision taken by a combination of multiple classifiers is better than the decision of only one classifier regardless of the strength of this unique classifier. In [16], the advantages of combining multiple classifiers or what is called a multiple classifier system are given. Face recognition, as one of the pattern recognition problems, is considered to be one of the most important fields especially after the 11th of September 2001 events. The need to automatically recognize the people from their faces becomes much imperative than before. In this paper, we introduce a new approach based on combining multiple classifiers instead of depending on one classifier to achieve better classification results. In Section 2, a literature review is given to explain the used techniques in combining the individual classifiers. Section 3 describes the Learning Vector Quantization (LVQ) neural network as a base classifier and explains the proposed approach. In Section 4, the different image databases used in this research like Yale, ORL and Essex are described. In Section 5, in addition to discussing the implementation of the proposed approach, comparisons against other different approaches are presented. Finally, a conclusion about the proposed approach and its ability to reach the objective of this paper is discussed.
Section snippets
Background and literature review
In pattern recognition literature, it has been shown that combining classifiers gives better results than individual classifiers. Some of these studies are in the areas of word recognition [16], facial-gender recognition [45] and face recognition [60]. In [30], mainly four groups that characterize the combination of the individual classifiers (or the Multiple Classifier System) have been identified: the representation of the input, the architecture of the individual classifiers, the
The proposed approach
In the proposed approach we make a combination of individual neural networks. The type of all individual neural networks is unified, but they are different in the main parameters of that type.
Face databases
To fairly evaluate the proposed combined classifier, a variety of research databases were used, which are, Yale database [65], ORL database [3] and Essex database [57]. Essex database, which includes images of persons of different racial origins, has four databases: Grimace [56], Faces94 [53], Faces95 [54], Faces96 [55]. Eventually, we used five databases: Yale, ORL, Grimace, Faces94 and Faces95 databases. Table 1 summarizes the main attributes of the original images of these databases:
Before
Case study: applying the proposed approach to Yale database
In the first step, we trained many individual classifiers. Each classifier is different than the others in the number of epochs, the learning rate, and/or the number of hidden neurons. With Yale database, we found that the suitable mixture of the values of the main parameters that gave us correct recognition accuracy of the training between 80% and 90% (or around these boundaries) are the following values: (1) number of epochs is between 300 and 400, (2) learning rate is between 0.02 and 0.06,
Conclusions
This paper has introduced a novel training weighted classifier ensemble (TWE), to solve the instability problem of individual pattern classifiers. The proposed classifier consists of multiple individual classifiers which are different in both architecture and training parameters. The achieved classification accuracies of the combined classifier outperform those of the best individual classifiers. For fair evaluation of the proposed combined classifier, the most widely used five face databases
Acknowledgement
The authors would like to acknowledge the support of their respective universities.
References (72)
- et al.
Face recognition using independent component analysis and support vector machines
Pattern Recognition Letters
(2003) - et al.
Error-correcting output coding corrects bias and variance
- et al.
Recognizing 2-tone images in grey-level parametric eigenspaces
Pattern Recognition Letters
(2002) - et al.
A complete fuzzy discriminant analysis approach for face recognition
Applied Soft Computing
(2010) - et al.
Sparsity preserving projections with applications to face recognition
Pattern Recognition
(2010) - et al.
A pair-wise decision fusion framework: recognition of human faces
- et al.
Error reduction through learning multiple descriptions
Machine Learning
(1996) - AT&T Laboratories Cambridge, The ORL Face Image Database, 2002, URL:...
- et al.
Voting schemes for cooperative neural network classifiers
- et al.
Democracy in neural nets: voting schemes for classification
Neural Networks
(1995)
Bagging predictors
Machine Learning
A simple implementation of the stochastic discrimination for pattern recognition
Machine learning research: four current directions
The AI Magazine
Experiments with classifier combining rules
Fundamentals of Neural Networks: Architectures, Algorithms, and Applications
A decision-theoretic generalization of on-line learning and an application to boosting
SIFT features for face recognition
Neural network ensembles
IEEE Transactions on Pattern Analysis and Machine Intelligence
Combination of decisions by multiple classifiers
Data complexity analysis for classifier combination
A boosting approach to P300 detection with application to brain–computer interfaces
Integrating faces and fingerprint for personal identification
IEEE Transactions on Pattern Analysis and Machine Intelligence
Face recognition by discriminative orthogonal rank-one tensor decomposition
A hybrid face recognition method using Markov random fields
International Conference on Pattern Recognition
Statistical pattern recognition: a review
IEEE Transactions on Pattern Analysis and Machine Intelligence
Combinations of weak classifiers
IEEE Transactions on Neural Networks
Eigenfeature regularization and extraction in face recognition
IEEE Transactions on Pattern Analysis and Machine Intelligence
Face recognition using support vector machines with local correlation kernels
International Journal of Pattern Recognition and Artificial Intelligence
On combining classifiers
IEEE Transactions on Pattern Analysis and Machine Intelligence
Stochastic discrimination
Annals of Mathematics and Artificial Intelligence
Self-organizing formation of topologically correct feature maps
Biological Cybernetics
Combining different methods and numbers of weak decision trees
Pattern Analysis and Applications
Face recognition: a convolutional neural-network approach
IEEE Transactions on Neural Networks
Face recognition using improved pairwise coupling support vector machines
Cited by (11)
Robust face descriptor in unconstrained environments
2024, Expert Systems with ApplicationsSoftly combining an ensemble of classifiers learned from a single convolutional neural network for scene categorization
2018, Applied Soft Computing JournalCitation Excerpt :Obtained scene features from an example image are illustrated in Fig. 6. It has been demonstrated numerous times that ensemble methods can usually yield superior results to single classifiers [65–67]. To make the ensemble methods effective, a diverse set of classifiers are necessary.
Generalizing intersection kernel support vector machines for color texture based recognition
2016, Journal of Visual Communication and Image RepresentationCitation Excerpt :Image texture provides the information about the spatial arrangement of color or intensities in an image or selected region of the image, which has the advantage of rotational invariance and stronger robustness to noise [4,5,7]. From the view of information fusion [8], a stronger feature can be constructed adopting certain fusion strategy from several heterogeneous features by use of their complementarity and redundancy. Previous studies [9,10] have shown that among various fusion schemes for bottom-layer image features, the one fusing color and texture has the attraction of strong discriminative power.
Combining features of negative correlation learning with mixture of experts in proposed ensemble methods
2012, Applied Soft Computing JournalCitation Excerpt :There are also hopes that if a classifier fails, the overall system can recover the error [3]. Combining methods [4,5] is an approach to improve the performance in prediction [6,7] and classification [8,9] particularly for complex problems such as those involving limited number of patterns, high-dimensional feature sets, and highly overlapped classes [10,11]. In this research, we focus on the combining methods in which NNs are used as the base classifier in the combining system.
Reward-Penalty Weighted Ensemble for Emotion State Classification from Multi-Modal Data Streams
2022, International Journal of Neural SystemsIndoor scene recognition based on weighted voting schemes
2019, 2019 European Conference on Mobile Robots, ECMR 2019 - Proceedings