Face Recognition Using Convolutional Neural Network and Simple Logistic Classifier
In this paper, a hybrid system is presented in which a convolutional neural network (CNN) and a Logistic regression classifier (LRC) are combined. A CNN is trained to detect and recognize face images, and a LRC is used to classify the features learned by
- PDF / 572,273 Bytes
- 11 Pages / 439.37 x 666.142 pts Page_size
- 73 Downloads / 220 Views
Abstract In this paper, a hybrid system is presented in which a convolutional neural network (CNN) and a Logistic regression classifier (LRC) are combined. A CNN is trained to detect and recognize face images, and a LRC is used to classify the features learned by the convolutional network. Applying feature extraction using CNN to normalized data causes the system to cope with faces subject to pose and lighting variations. LRC which is a discriminative classifier is used to classify the extracted features of face images. Discriminant analysis is more efficient when the normality assumptions are satisfied. The comprehensive experiments completed on Yale face database shows improved classification rates in smaller amount of time.
1 Introduction Effective methods in the extraction of features and classification methods for the extracted features are the key factors in many real-world pattern recognition and classification tasks [1]. Neural networks such as multilayer perceptron (MLP) are considered as one of the simplest classifiers that can learn from examples. An MLP can approximate any continuous function on a compact subset to any desired accuracy [2].
H. Khalajzadeh (&) M. Mansouri M. Teshnehlab K. N. Toosi University of Technology, Tehran, Iran e-mail: [email protected] M. Mansouri e-mail: [email protected] M. Teshnehlab e-mail: [email protected]
V. Snášel et al. (eds.), Soft Computing in Industrial Applications, Advances in Intelligent Systems and Computing 223, DOI: 10.1007/978-3-319-00930-8_18, Springer International Publishing Switzerland 2014
197
198
H. Khalajzadeh et al.
The performance of such classifiers depends strongly on an appropriate preprocessing of the input data [3]. In the traditional models a hand designed feature extractor congregates relevant information from the input. Designing the feature extractor by hand requires a lot of heuristics and, most notably, a great deal of time [4]. Moreover, it is not apparent what illustrates an optimal preprocessing or if there even exists an optimal solution. Sometimes it is required the input dimension to be reduced as far as possible [3]. The main deficiency of MLP neural networks for high dimensional applications such as image or speech recognition is that they offered little or no invariance to translations, shifting, scaling, rotation, and local distortions of the inputs [4]. Convolutional neural networks [5] were proposed to address all problems of simple neural networks such as MLPs. CNNs are feed-forward networks with the ability of extracting topological properties from the unprocessed input image without any preprocessing needed. Thus, these networks integrate feature selection into the training process [3]. Furthermore, CNNs can recognize patterns with extreme variability, with robustness to distortions and simple geometric transformations like translation, scaling, rotation, squeezing, stroke width and noise [6, 7]. Different versions of convolutional neural networks are proposed in the literature.
Data Loading...
