Shadow determination and compensation for face recognition
- PDF / 431,893 Bytes
- 7 Pages / 595.276 x 790.866 pts Page_size
- 62 Downloads / 157 Views
ORIGINAL ARTICLE
Shadow determination and compensation for face recognition Yi Li • Xiaodong Liu • Zhijun Gao
Received: 18 June 2013 / Accepted: 8 October 2013 Ó Springer-Verlag Berlin Heidelberg 2013
Abstract Illumination variation that occurs on face images will significantly influence the performance of the recognition. Based on the low-dimensional intrinsic of face images, we design a novel two-step shadow compensation method for face recognition. Three indexes are proposed and employed to distinguish the shaded-images. Then, we compensate the shadows adaptively by using a modified Robust PCA result. Experimental results on Yale database and Yale B database demonstrate that the proposed approach can improve the recognition rate. Showing our method is suitable for face recognition with illumination variations. Keywords Face recognition Illumination variation Shadow compensation Robust PCA Low-dimensional
1 Introduction Recently, there are extensive researches on face recognition following its increased applications in various fields such as security surveillance, general identity verification and so on. The well-known approaches include Eigenface [1], linear discriminant analysis (LDA) [2] and their
Y. Li (&) X. Liu School of Electronic and Information Engineering, Dalian University of Technology, Dalian 116024, People’s Republic of China e-mail: [email protected] Y. Li Z. Gao Department of Computer and Information Engineering, Heilongjiang University of Science and Technology, Harbin 150027, People’s Republic of China
extensions such as 2DLDA method [3], which are demonstrated their success under controlled circumstances. However, face images are significantly affected by varying lighting condition. The performance of face detection algorithms are severely affected by the illumination variations [4]. Also, it is a major cause of accuracy degraded of the face recognition in many practical applications [5]. The algorithms about how to remove the illumination variation have been received much attention in recent years. Many image enhancement techniques are used for face recognition. For example, principal component analysis based face image enhancement technique [6], histogram equalization (HE) and its modified techniques such as block-based histogram equalization (BHE) have been introduced to cope with this issue [7]. Nevertheless, [8] showed that histogram equalization represents only a specific case. Nonuniform illumination variation is still difficult to deal with using the global processing techniques [9]. Based on facial symmetry and image average, Hsieh et al. [10] suggested a compensation approach, which can overcome the drawbacks of histogram equalization [11, 12] achieved illumination invariance through the illumination reflection model. Lee et al. [13] proposed the 9D linear subspace approach by using nine images captured under nine lighting directions. However, these methods need a large number of samples to learn the varied lighting, that is, they require sufficient numbers of trainin
Data Loading...
