基于双空间金字塔匹配核图像目标分类.pdfVIP

403 Vol. 40, N o. 3 2 0 10 3 JOURNAL OF UNI VERSIT Y OF SCIENCE AND T ECHNOLOGY OF CHINA Mar. 2 0 1 0 :0253-2778(2010) 03-0313-08 陈海林, 吴秀清 ( , 230027) : 提出一种基于局部特征的双空间金字塔匹配核( bi-space pyramid match kernel, BSPM) 用于 图像目标分类. 利用局部特征在特征空间和图像空间建 统 一的多分辨率框架, 以便较好地表达图 像的语义内容. 该方法同时在特征空间和图像空间建 金字塔型结构, 通过适当匹配可以得到正定 核函数, 该函数具有线性计算复杂度, 可以运用于基于核的学习算法. 将BSPM 嵌入支持向量机对 公共数据库中图像目标进行分类, 实验结果表明该方法对图像具有良好的分类能力, 优于词汇导向 的金字塔匹配核和空间金字塔匹配核. : 双空间金字塔匹配核; 局部特征; 空间金字塔; 特征空间金字塔 : T P391. 41 : A doi: 10. 3969/ j. issn. 0253-2778. 2010. 03. 018 B-i space pyramid match kernel for image object classification CHEN Hailin, WU Xi qing (Dep artment of Electronic Engineer ing and I nf ormation Science, University of Science and Technology of China, H ef ei 230027, China) Abstract: Local feat res based b-i space pyramid match kernel ( BSPM) w as proposed for image object classification. In both feat re space and image space, a nified m lti-resol tion framew ork w as b ilt sing the local feat res in order to better describe the semantic content of an image. T he pyramids can be b ilt in both the feat re space and image space w ith the proposed method, and the positive definite kernel f nction can be obtained thro gh s itable matching. This f nction w ith linear comp tation cost can be sed in kerne-l based learning algorithms. T he BSPM w as embedded into the s pport vector machine for image classification on the p blic image databases, and the experiments show that the proposed method can achieves good classification ability and o tperform both vocab lary-g ided pyramid match kernel and spatial pyramid match kernel. Key words: b-i space pyramid match kernel; local feat re; spatial pyramid; pyram