Face recognition with Symmetric Local Graph Structure (SLGS)
Introduction
Face recognition is an important area in biometrics and computer vision. It has a wide range of applications in information security, image retrieval, access control and law enforcement surveillance. The area of face recognition has achieved a significant research progress. This is due to the current systems perform well under relatively controlled environments but tend to suffer when variations in different factors, such as aging, illumination, facial expressions and pose, are present. Thus the main goal of the ongoing research is to increase the performance and robustness of the system against various factors.
A critical survey of face recognition research had been conducted by Zhao, Chellappa, Phillips, and Rosenfeld (2003). The survey categorises existing recognition techniques and provides detailed descriptions of representative methods within each category. Three major categories of face recognition from intensity images are holistic, feature-based and hybrid. Holistic approach utilises the information derived from the whole face images. It generates a general template for the whole face pattern. The recognition will be based on this general template. Among the major approaches in this category are Principal Component Analysis (PCA) (Turk & Pentland, 1991), Linear Discriminant Analysis (LDA) (Etemad & Chellappa, 1997), and Independent Component Analysis (ICA) (Bartlett, Movellan, & Sejnowski, 2002). Another category, called feature-based, compares the salient facial features or components detected from the face. Examples of approaches in this category are Local Graph Structure (LGS) (Sayeed, Yusof, Bashier, & Hossen, 2013), Local Binary Patterns (LBP) (Ojala, Pietikäinen, & Harwood, 1996), and Elastic Bunch Graph Matching (EGBM) (Wiskott, Fellous, Krugerl, Malsburg, & Vin Der, 1997). The last category that is hybrid approach combines both local and global features to produce a more complete facial representation (Liu and Chen, 2007, Liu and Liu, 2010, Mandal and Dhara, 2009).
In this work, a new approach for face recognition based on LGS is presented. The idea of LGS is to represent a pixel in an image with a graph structure of its neighbours’ pixels in order to capture the spatial information. The LGS operator extracts more information about the texture of the face images compared to LBP. Thus it produces better accuracy in face recognition problem (Abusham & Bashir, 2011). However, the proposed graph structure of LGS was non-symmetric as in Fig. 2. The graph structure of LGS represents more left-handed neighbour pixels then the right-handed. Thus, this work proposed Symmetric Local Graph Structure (SLGS) with a symmetric graph structure. This approach gives a better representation of local features for each pixel. As the result, the proposed technique produces better performance for face recognition.
Section snippets
Local binary patterns
Local Binary Patterns (LBP) operator has been introduced by Ojala, Pietikainen, and Harwood (1994). It is a simple yet very efficient and powerful operator to describe textures. The operator labels the pixels of an image by thresholding the 3 × 3-neighbourhood of each pixel with the centre value and considers the result as a binary number as shown in Fig. 1. LBP operator has become a popular approach in various applications, including face recognition (Ahonen, Hadid, & Pietikkainen, 2004),
Experimental design
A publicly available AT&T (formerly known as ORL) and Yale face databases were utilised to test the performance of SLGS. AT&T face database contains 400 images for 40 distinct subjects. The images were taken at different times, lighting, facial expressions (open/closed eyes, smiling/not smiling) and facial details (glasses/no glasses). All the images were taken against a dark homogeneous background with the subjects in an upright, frontal position. Fig. 6 shows a part of the preview images in
Results
First experiment was conducted to compare the effectiveness of three different distance measures which are Euclidean distance, correlation coefficient and chi-square statistics. The experiment used only the LBP for feature extraction algorithm on the AT&T database. Table 1 shows the result of the experiment with different number of training files. Chi-square distance measure gave highest accuracy in all different number of training files. However, the difference of accuracy given by different
Discussion and conclusion
Face images can be illustrated as a collection of micro-patterns which can be well described by LBP. It is then being improved with the graph representation in LGS. We extended the idea of the graph representation and proposed a symmetric graph structure called SLGS. In this approach, the graph structure of a pixel in an image has better representation with its neighbours’ pixel. Face recognition is performed using a nearest neighbour classifier in the computed feature space with Euclidean
References (24)
- et al.
Fusion of color, local spatial and global frequency information for face recognition
Pattern Recognition
(2010) - et al.
A comparative study of texture measures with classification based on feature distributions
Pattern Recognition
(1996) - et al.
Integrated multilevel image fusion and match score fusion of visible and infrared face images for robust face recognition
Pattern Recognition
(2008) - et al.
Face recognition using local graph structure (LGS)
Lecture Notes in Computer Science
(2011) - et al.
Face recognition with local binary patterns
Lecture Notes in Computer Science
(2004) - et al.
A hierarchical face recognition method based on local binary pattern
- et al.
Face recognition by independent component analysis
- et al.
Face detection based on graph structure and neural networks
Trends in Applied Sciences Research
(2012) - et al.
Discriminant analysis for recognition of human face images
Journal of the Optical Society of America
(1997) - et al.
Face detection using improved LBP under bayesian framework