Reducing JointBoost-based multiclass classification to proximity search
Miami, FL(2009)
摘要
Boosted one-versus-all (OVA) classifiers are commonly used in multiclass problems, such as generic object recog- nition, biometrics-based identification, or gesture recog ni- tion. JointBoost is a recently proposed method where OVA classifiers are trained jointly and are forced to share fea- tures. JointBoost has been demonstrated to lead both to higher accuracy and smaller classification time, compared to using OVA classifiers that were trained independently and without sharing features. However, even with the im- proved efficiency of JointBoost, the time complexity of OVA- based multiclass recognition is still linear to the number o f classes, and can lead to prohibitively large running times in domains with a very large number of classes. In this pa- per, it is shown that JointBoost-based recognition can be re - duced, at classification time, to nearest neighbor search ina vector space. Using this reduction, we propose a simple and easy-to-implement vector indexing scheme based on princi- pal component analysis (PCA). In our experiments, the pro- posed method achieves a speedup of two orders of magni- tude over standard JointBoost classification, in a hand pose recognition system where the number of classes is close to 50,000, with negligible loss in classification accuracy. Ou r method also yields promising results in experiments on the widely used FRGC-2 face recognition dataset, where the number of classes is 535.
更多查看译文
关键词
computational complexity,image classification,learning (artificial intelligence),principal component analysis,search problems,JointBoost-based multiclass classification,OVA-based multiclass recognition,nearest neighbor search,one-versus-all classifier,principal component analysis,proximity search,time complexity,vector indexing,vector space
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络