A Comparative Study of Invariant Range Image Multi-Pose Face Recognition Using K-Means, Fuzzy C-Means, Membership Matching Score and Center of Gravity Search

摘要

This paper is proposed for analogy of efficiency of the clustering algorithm and the method to search the appropriate pose position for matching in invariant range image multi-pose face recognition system. The center of gravity search is used for searching pose position in range image face database (RIFD). Reference persons data base are grouped by using K-means compared with Fuzzy c-means cluster for speed up processing time. This approach is developed for implementation the invariant range image multi-pose face recognition system. This face recognition system is created to function covering pose variation region ±24 degrees up/down and left/right (UDLR) from initial pose. RIFD used in this face recognition is based on 3-D Graphics database. For this advantage, we could solve scale, center and pose error problem by using geometric transform. RIFD that is obtained from range image sensors will be used for operation by reducing data size. RIFD will be transformed by the gradient transform into significant feature and matching by using membership matching score. The proposed method was tested using facial range images from 130 persons with normal facial expressions. The processing time of the recognition system has to be better than 3LMS by the speeding up to 10 times without any change of recognition rate. The output of the detection and recognition system has to be accurate to about 92 percent.