视觉字典合理容量的自动获取研究.pdfVIP

2014 年 9 月 第 35 卷第 9期 计算机工程与设计 Sept. 2014 Vol. 35 No. 9 COMPUTER ENGINEER卧~G AND DESIGN 视觉字典合理容量的自动获取研究 钟映春，详志+， 孙伟，连伟烯 (广东工业大学自动化学院，广 东广 州 510006) 摘 要:针对当前完全依赖反复人为实验摸索才能获得视觉字典容爱的现状，提必一种自动计算视觉字典合理容量的方 法。采用尺度不交特征转换 (SIFT) 局在11描述子构建场景图像数据集的特征矩碎，采用主成分分析 (PCA) 方法进行降维 处理;采用吸引子传播聚类方法 (AP 聚类〉对特征矩阵进行聚类处理，估算获得视觉字典的合理容量，并将合理容量的 视觉字典用于支持向萤机 (SVM) 中进行场景分类的训练和识别。将该方法与级决的 K-means视觉字典容量获取方法选行 对比，对比给采表明，该方法提高了场景分类的粘度，显著减少了场景分类的运行时间，提高了计算效率。 关键词: 视觉字典容量; SIFT描述子;主成分分析;AP 聚类算法;支持向盘机 中图法分类号: TP391. 41 文献标i只号 A 文章编号; 1000-7024 (2014) 09-3279-05 Research on reasonable capacity of visual dictionary obtaining ZHONG Ying-chun , TAN Zhi+ , SUN W ei , LlAN W ei-xi (School of Automation, Guangdong Ulúversity of T echnology , Guangzhou 510006 , China) Abstract; For the situation that the visllal dictionary capacity is obtained through artificial experiment repeatedly , a method was proposed that calclllated the reasonable capacíty of the viSllal dictionary automatically. F让st ， the scale-invariant featUTe transform (SIFT) local feature descriptors were employed ωωnstruct the feature matrix of the scene image data set, whose dimension was reduced by using principal component analysis (PCA) . Second , the feature matrix was clust.ered through using aIfinity propaga- tion 仙的 method to obtain the reasonable capacity of 叩ual dictionary. And the reasonable capacity of visual dictionary was employed t.o tr♂n support vector machine (SVM) and scene t.YI咒 recognition. Finally, the proposed method was compa.red with 出e K-means which was tlle c1a岱ic method for the acquisition of visual dictionaryωpacity. Expelimenal results show tllat the proposed method not on1y im- proves the accul-acy of the scene type re∞gnition ， but reduces the running time and increases the ∞mputation efficiency. Kcy words: capacity of the visual dictionary; SIF寸 descriptors; pr切cipal ∞mponent analysis; AP clustering; sup阳此 vecωr maclúne 。引言 具有合理容在的视觉字典可以有效提高场景分类的精 度和识别效率。 自动获取视觉字典容量可以有效避免通过 大量实验去寻找合理的视觉字典容量，进而显著提高算法 的