Face Recognition Based on Enhanced CSLBP
Face image with illumination variation usually contains redundant data that will seriously reduce the recognition rate. To combat the influence of illumination variation and extracting illumination-robust feature, a novel feature extraction method is prop
- PDF / 612,264 Bytes
- 6 Pages / 439.37 x 666.142 pts Page_size
- 89 Downloads / 208 Views
2
1 College of Computer Science, North China University of Technology, Beijing, China [email protected] Beijing Key Laboratory on Integration and Analysis of Large-Scale Stream Data, Beijing, China
Abstract. Face image with illumination variation usually contains redundant data that will seriously reduce the recognition rate. To combat the influence of illumination variation and extracting illumination-robust feature, a novel feature extraction method is proposed. The novel method is based on the combination of Center-Symmetric Local Binary Pattern (CS-LBP) and the fusion of the vertical and horizontal component images derived from wavelet decomposition. Numerous experiments have been done on the Extended Yale B to verify its effectiveness. The experimental results show that by applying the proposed method, redundant data caused by severe illumination variation can be filtered, while useful texture information can be reserved and enhanced. Compared with CSLBP, it significantly improves the face recognition performance under severe illumination variation. Keywords: Wavelet decomposition extraction
Image fusion
CSLBP
Feature
1 Introduction Face recognition has been a research hotspot ranging from the field of biometric recognition to modern security application due to its’ distinct advantage compared with other biometric characteristics. Abundant face recognition approaches have been proposed, such as Center Symmetric Local Binary Pattern (CSLBP) [1–3], Local Directional Binary Pattern (LDBP) [4] and so on. Most of them can achieve satisfactory result if the face images are captured under well-constrained circumstances. However, in real scenario, there are many unconstrained factors such as illumination variation, expression variation, pose variation or occlusion that can weaken the face recognition performance. To combat these influence factors, further explores are urgent. This paper focuses on study novel feature extraction method that can be more robust to illumination variation. Besides the rich information for identification, natural face image with illumination variation also contains redundant data that can lead to appearance variation of the image. To combat the variation and lower the impact of the redundant data, various pre-processing algorithms such as histogram equalization [5] and wavelet transform [6] have been applied. However, some useful information for recognition can also be suppressed during the preprocessing. Another category of method to combat illumination © Springer Nature Singapore Pte Ltd. 2017 J.J. (Jong Hyuk) Park et al. (eds.), Advanced Multimedia and Ubiquitous Engineering, Lecture Notes in Electrical Engineering 448, DOI 10.1007/978-981-10-5041-1_86
540
C. Li et al.
variation is feature based method. Researchers try to propose feature which is more robust to illumination variation, such as Local Binary Pattern (LBP) [7], CSLBP, Gabor wavelet [8, 9] and so on. Among these, CSLBP has been proven to be efficient at describing the texture features of image. However its robu
Data Loading...
