点击次数：

论文类型：期刊论文

发表时间：2010-03-01

发表刊物：Journal of Information and Computational Science

收录刊物：Scopus、EI

卷号：7

期号：3

页面范围：637-648

ISSN号：15487741

摘要：Recognition accuracy and classification speed are two essential issues in face recognition. Approaches using Gabor filtering are effective in improving recognition accuracy. For instance, the approach based on Local Gabor Binary Pattern Histogram Sequence (LGBPHS) has been applied to face recognition successfully, and achieves high recognition rate. However, it suffers from excessive computational complexity, which limits its applications. To increase the computational speed, this paper proposes the parallel LGBPHS and weighted LGBPHS approaches. Many simulation experiments have been carried out to test the performance of the parallel algorithms on the ORL face database and the FERET face database by using a cluster of 10 homogeneous dual-processor PCs. The experimental results show that the parallel algorithms presented in the paper are effective in speeding up the algorithms of training and classification, while maintaining the recognition accuracy unchanged. Besides, the parallel performance of the algorithms has been further improved with the increase in the size of face database, which shows the good scalability of the parallel algorithms. Copyright ? 2010 Binary Information Press.