针对传统人脸识别方法实时性差的缺点,提出了一种加速鲁棒性特征(SURF,speed up robust features)和词包模型(BOW,bag-of-word)相结合的人脸识别方法.图像经过预处理后,使用SURF算法自动提取图像的关键点和相应的特征描述符,再进一步用BOW方法将其编成视觉单词作为人脸的局部特征.最后,采用K最邻近结点算法进行分类识别.使用了2个数据集验证了提出的方法——标准CMU-PIE(卡内基梅隆大学——姿势、光照、表情人脸数据库)人脸库和采集的数据库,分别达到了97.5%和99.3%的识别率,而且特征提取的时间少于0.108 s,识别的时间少于0.017 s.结果表明,本文提出的算法不仅精确而且快速,具有更好的稳定性和有效性.
To overcome the limitations of traditional face recognition methods for real-time, a face recognition method which based on speed up robust features and bag-of-word model was proposed.Image after preprocessing, we used SURF to extract key points of images and corresponding feature descriptors automatically.Further, bag-of word model was used to code the descriptors into visual words as local features of the face.Finally, K-Nearest Neighbor algorithm was adopted to recognize the human faces.The proposed method is validated with both CMU-PIE dataset and dataset collected in the laboratory.It can achieve 97.5% and 99.3% recognition rates on these two datasets, respectively.In average, it took less than 0.108 s for feature extraction and less than 0.017 s for matching.The results indicate that the proposed method not only precise moreover fast, and had better stability and effectiveness.
[1] SOLDERA J, CARLOS A R B, SCHARCANSKI J.Customized orthogonal locality preserving projections with soft Margin maximization for face recognition[J].IEEE Transactions on Instrumentation & Measurement, 2015, 64(9):2417-2426.DOI:10.1109/TIM.2015.2415012.
[2] ZHANG W Y, HUA L L, DU H R, et al. SIFT: predicting amino acid changes that affect protein function[Z]. International Symposium on Computer Network and Multimedia Technology, Wuhan, 2009.
[3] LOWE D G.Object recognition from local scale-invariant features [J].Computer Vision, 1999, 2(9):1150-1157.DOI:10.1109/ICCV.1999.790410.
[4] LIU C, YUEN J, TORRALBA A.SIFT flow: Dense correspondence across scenes and its applications[J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 2011, 33(5):978-994.DOI:10.1109/TPAMI.2010.147.
[5] GENG C, JIANG X D.Face recognition based on the multi-scale local image structures[J].Pattern Recognition, 2011, 44(10):2565-2575.DOI:10.1016/j.patcog.2011.03.011.
[6] 李伟红, 陈龙, 龚卫国.单样本快速人脸不变特征提取方法[J].光电子·激光, 2014(3):558-564. LI W H, CHEN L, GONG W G.Fast facial descriptor extraction for single image based face recognition[J].Journal of Optoelectronics·Laser,2014(3):558-564.
[7] DU G, SU F, CAI A.Face recognition using SURF features[J].Pattern Recognition,2009,7496(28):131-137.DOI:10.1117/12.832636.
[8] BAY H, ESS A, TUYTELAARS T, et al.Speeded-up robust features(SURF)[J].Computer Vision and Image Understanding, 2008, 110(3):346-359.DOI:10.1016/j.cviu.2007.09.014.
[9] DING N N, LIU Y Y, ZHANG Y, et al.Fast image registration based on SURF-DAISY algorithm and randomized kd trees[J].Journal of Optoelectronics Laser, 2012, 23(7):1395-1402.
[10] KANG M, CHOO W, MOON S. Improved face recognition algorithm employing SURF descrpitors[Z]. Society of Instrument and Control Engineers Annual Conference, Japan,2010.
[11] LEWIS D D.Naive(Bayes)at forty: The independence assumption in information retrieval[J].European Heart Journal, 1998, 21(15):1204-6.
[12] LI F F, PERONA P. A Bayesian hierarchical model for learning natural scene categories[Z]. Computer Society Conference on Computer Vision & Pattern Recognition, America, 2005.
[13] YANG J, JIANG Y G, HAUPTMANN A G, et al. Evaluating bag-of-isua-words representations in scene classification[Z].International Workshop on Workshop on Multimedia Information Retrieval, ACM, 2007. DOI:10.1145/1290082.1290111
[14] CHU W T, HUANG C C, CHENG W F. News story clustering from both what and how aspects: using bag of word model and affinity propagation[Z]. The 2011 ACM International Workshop on Automated Media Analysis and Production for Novel TV Services, ACM, 2011.
[15] FENG J, JIAO L C, ZHANG X, et al.Bag-of-Visual-Words based on clonal selection algorithm for SAR image cassification[J].Geoscience & Remote Sensing Letters IEEE, 2011, 8(4):691-695.DOI:10.1109/LGRS.2010.2100363.
