Local Zernike Moment and Multiscale Patch-Based LPQ for Face Recognition
In this paper, a novel feature extraction method combining Zernike moment with multiscale patch-based local phase quantization is introduced, which can deal with the problem of uncontrolled image conditions in face recognition, such as expressions, blur,
- PDF / 247,687 Bytes
- 9 Pages / 439.37 x 666.142 pts Page_size
- 35 Downloads / 183 Views
Abstract In this paper, a novel feature extraction method combining Zernike moment with multiscale patch-based local phase quantization is introduced, which can deal with the problem of uncontrolled image conditions in face recognition, such as expressions, blur, occlusion, and illumination changes (EBOI). First, the Zernike moments are computed around each pixel other than the whole image and then double moment images are, respectively, constructed from the real and imaginary parts. Subsequently, multiscale patch-based local phase quantization descriptor is utilized for the non-overlapping patches of moment images to obtain the texture information. Afterward, the support vector machine (SVM) is employed for classification. Experimental results performed on ORL, JAFFE, and AR databases clearly show that the LZM-MPLPQ method outperforms the state-of-the-art methods and achieves better robustness against severe conditions abovementioned. Keywords Local Zernike moment EBOI
⋅ Local phase quantization ⋅
Face recognition
⋅
1 Introduction With the wide applications of surveillance, automation, and intelligent devices, face recognition (FR) has become one of the most active research areas of computer vision. Even though there are many methods proposed in FR [1–3] during the last X. Sun ⋅ X. Fu (✉) ⋅ Z. Shao ⋅ Y. Shang ⋅ H. Ding College of Information and Engineering, Capital Normal University, Beijing 100048, China e-mail: [email protected] X. Fu ⋅ Z. Shao ⋅ Y. Shang ⋅ H. Ding Beijing Advanced Innovation Center for Imaging Technology, Capital Normal University, Beijing 100048, China Y. Shang ⋅ H. Ding Beijing Key Laboratory of Electronic System Reliability Technology, Capital Normal University, Beijing 100048, China © Springer Science+Business Media Singapore 2016 Y. Jia et al. (eds.), Proceedings of 2016 Chinese Intelligent Systems Conference, Lecture Notes in Electrical Engineering 405, DOI 10.1007/978-981-10-2335-4_3
19
20
X. Sun et al.
two decades, it is still a challenging problem to adapt different conditions such as expressions, blur, occlusion, and illumination changes (EBOI). Generally, a FR system is constituted of feature extraction and pattern classification. It is of great significance to choose proper method to extract features. Global and local features are two classes of feature extraction methodologies. Since local features exemplified by LBP [4] and Gabor [5] have better affine invariance and illumination insensitivity compared with global features, they are usually applied to the FR system with changeable image variations. However, LBP gets poor antinoise ability while Gabor method suffers from high computational cost because of processing 40 components (5 scales and 8 orientations) for every facial image. In recent years, many other local feature methods have been proposed to achieve better performance. For example, Ojansivu and Heikkila [6] presented the local phase quantization method (LPQ) in discrete Fourier transform domain. To further improve the performance, LPQ family is enriched by patch-bas
Data Loading...
