摘要
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立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立金字塔型结构,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.
Local features based bi-space pyramid match kernel（BSPM） was proposed for image object classification.In both feature space and image space,a unified multi-resolution framework was built using the local features in order to better describe the semantic content of an image.The pyramids can be built in both the feature space and image space with the proposed method,and the positive definite kernel function can be obtained through suitable matching.This function with linear computation cost can be used in kernel-based learning algorithms.The BSPM was embedded into the support vector machine for image classification on the public image databases,and the experiments show that the proposed method can achieves good classification ability and outperform both vocabulary-guided pyramid match kernel and spatial pyramid match kernel.
