An innovative face image enhancement based on principle component analysis
- PDF / 706,686 Bytes
- 9 Pages / 595.276 x 790.866 pts Page_size
- 87 Downloads / 222 Views
ORIGINAL ARTICLE
An innovative face image enhancement based on principle component analysis Xiang Xu • Wanquan Liu • Svetha Venkatesh
Received: 18 June 2011 / Accepted: 14 November 2011 / Published online: 7 December 2011 Ó Springer-Verlag 2011
Abstract In this paper, we propose an innovative face hallucination approach based on principle component analysis (PCA) and residue technique. First, the relationship of projection coefficients between high-resolution and low-resolution images using PCA is investigated. Then based on this analysis, a high resolution global face image is constructed from a low resolution one. Next a highresolution residue is derived based on the similarity between the projections on high and low resolution residue training sets. Finally by combining the global face and residue in high resolution, a high resolution face image is generated. Also the recursive and two-stage methods are proposed, which improve the results of face image enhancement. Extensive experiments validate the proposed approaches. Keywords
PCA Super resolution Face hallucination
1 Introduction In surveillance systems, video frames are often in low resolution and they are difficult to use directly for some applications such as face recognition [37–45]. Super-resolution techniques [1–12, 19–36] are thus required. In general there are two kinds of super-resolution techniques. Interpolation technique increases the image resolution directly from the low-resolution input images [1–6], but does not achieve good results since it utilizes information
X. Xu W. Liu (&) S. Venkatesh Department of Computing, Curtin University, Perth, WA, Australia e-mail: [email protected]
only from low resolution images. Another technique is based on machine learning [7–9]. This framework includes two data sets: the low resolution and corresponding high resolution training samples. Since the purpose of image enhancement is to obtain a high resolution image for a corresponding low resolution image, learning algorithms aim to explore the relationship between high resolution and low solution images. Freeman et al. [7] develop a Markov Network to learn the relationship. Hertzmann et al. [8] apply the ‘Image Analogies’ method to obtain high resolution images using local feature transforms. However, these methods do not utilize the special features of face images and only perform well for textured images. For face image enhancement, Baker and Kanade [10] first propose the face hallucination algorithm. Based on principle component analysis (PCA) theory [13–18], which is widely used in computer vision, Wang and Tang [19] propose an eigen-transformation approach. They project the low resolution face image into the eigen-faces of the low-resolution training set and obtain the coefficients which can be used to construct a high resolution image. This method is computationally efficient for face image resolution enhancement. However only the global face is derived and thus some local features are not well characterized. For example, hair and glasses
Data Loading...
