A 3D Face Recognition Algorithm Based on Nonuniform Re-sampling Correspondence
This paper proposes an approach of face recognition using 3D face data based on Principle Component Analysis (PCA) and Linear Discriminant Analysis (LDA). This approach first aligned 3D faces based on nonuniform mesh re-sampling by computing face surface
- PDF / 489,221 Bytes
- 10 Pages / 430 x 660 pts Page_size
- 85 Downloads / 163 Views
Abstract. This paper proposes an approach of face recognition using 3D face data based on Principle Component Analysis (PCA) and Linear Discriminant Analysis (LDA). This approach first aligned 3D faces based on nonuniform mesh re-sampling by computing face surface curves. This step achieves aligning of 3D prototypes based on facial features, eliminates 3D face size information and preserves important 3D face shape information in the input face. Then 2D texture information and the 3D shape information are extracted from 3D face images for recognition. Experimental results for 105 persons 3D face data set obtained by Cyberware 3030RGB/PS laser scanner have demonstrated the performance of our algorithm.
1
Introduction
Face recognition technology have many potential application in public security, personal identification, automated crowd surveillance and so on. Over the past 40 years many different face recogniton techniques with 2D images were proposed. Although significant progress has been made [1], some difficult problems have not been solved well such as pose and illumination. The reason is that the input face image of same person is not taken at similar condition as that one of same person in the database. It has been proved [2,3] even small variations in pose and illumination can drastically degrade performance of 2D image based face recognition system. Pose and illumination invariant face recognition is a challenging research area. Some approaches have been proposed. A well known approach for achieving pose and illumination invariant is to utilize 3D face information [4]. As the technology for acquiring 3D face information becomes simpler and cheaper, the use of 3D face data becomes more common. A 3D face image contains its shape and texture information. 3D shape information which is lacking in 2D image is expected to provide more recognition feature and these features are robust against pose and illumination variations. So the use of additional 3D information is expected to improve the reliability of face recognition system. Principle Component Analysis (PCA) is a classical data compression, feature extraction and data representation technique widely used in the areas of pattern recognition. It has been one of the most successful approaches in face G. Bebis et al. (Eds.): ISVC 2007, Part II, LNCS 4842, pp. 407–416, 2007. c Springer-Verlag Berlin Heidelberg 2007
408
Y. Sun, J. Wang, and B. Yin
recognition [5]. Its goal is to find a basis vectors in an orthogonal linear space by training set and project a probe face image to the linear space spanned by the basis vectors. PCA can represent effectively a probe face image (not in the training set) as a linear combination of basis vectors. The different face image has different combination coefficients. These combination coefficients are taken as PCA feature for face recognition. It has been proved PCA transform is the optimal transform in the sense of minimum square error. But the correspondence of all face features is needed according to linear object class theory. In
Data Loading...
