毕业论文

当前位置: 毕业论文 > 计算机论文 >

直方图相交内核支持向量机改进分类方法

时间:2017-11-13 21:34来源:毕业论文
一个基于这些特征的相似直方图相交内核向量机分类器有目前最好的性能,在每个窗口 级别的误检下,仅仅有13%的漏检率,低于Dalal和Triggs的线性SVM探测器。在Daimler Chrysler行人数据集上
摘要用内核SVMs进行简单分类需要估计测试向量和支持向量的核函数。对于一类内核,本文展示能做的更有效。特别地,展示了能够建立直方图相交内核向量机(IKSVMs),其时间复杂度是现行的标准方法支持向量数目的对数级,而不是线性级。本文进一步展示,通过预先辅助表,能构造一个具有不变运行时间和空间需求的相似分类器。并且不依赖于支持向量的个数,而且在分类精度上几乎不变。这种近似法也取决于 和类似形式的其他内核。15085
同时也介绍了基于方向边缘势能的多层直方图的新特征,并展示了在多个探测数据集上的实验。在INRIA行人数据集上,一个基于这些特征的相似直方图相交内核向量机分类器有目前最好的性能,在每个窗口 级别的误检下,仅仅有13%的漏检率,低于Dalal和Triggs的线性SVM探测器。在Daimler Chrysler行人数据集上,直方图相交支持向量机(IKSVM)给出了和(基于最小二乘向量机的)最好结果相当的精准度,但速度却是最小二乘支持向量机的15倍。在这些实验中,相似直方图相交支持向量机的速度是一次标准实施速度的2000倍,却需要200倍更小的内存。最后展示了,用基于空间金字塔特征的相似直方图交叉内核支持向量机工作在Caltech 101数据集上,速度是原来的50倍,而精确度几乎不变。本文在有限的样本下,实现了这一功能,在实验部分给出了结果。
关键词  IKSVMs  方向边缘能量多层直方图 HOG SVM 分类精度 分类速度 相似       IKSVM
毕业设计说明书(论文)外文摘要
      Title    A new method of classification:the histogram interesection kernel SVMs
Abstract
   Concise classification using kernel SVMs need to estimate the kernel of a test vector and each support vector.The paper shows that the kernel can classify much more efficiently.Especially,shows that the paper establishes histogram intersection kernel SVMs (IKSVMs),its runtime complexity of the classifier is logarithmic of the number of support vectors for the current standard method,rather than linear. It further shows that trough the auxiliary tables in advance,one can construct an approximate classifier in changeless runtime and space,not rely on the support vectors number,and the classification accuracy is almost the same.This approximation depends on   and other similar form kernels.

源自六"维~论`文}网"加7位QQ3249.114 www.lwfree.cn


   At the same time ,it also introduces new features based on a multi-level histograms of oriented edge energy and some experiments on four detection datasets. On the INRIA pedestrian dataset, at present an approximate IKSVM classifier based on these features has the best performance,under   False Positive Per Window,with a miss rate 13% lower than the linear SVM detector of Dalal & Triggs.IKSVM gives comparable accuracy to the best results (based on quadratic SVM) of the Daimler Chrysler pedestrian dataset,with being 15× faster. In these experiments the approximate IKSVM is 2000× faster than a standard implementation and requires 200× less memory. Finally it shows that a 50× speedup is possible using approximate IKSVM based on spatial pyramid features with negligible loss of accuracy  on the Caltech 101 dataset.Under limited samples, this paper realized the function.In the part of experiments, results are given.
Keywords: IKSVMs  a multi-level histograms of oriented edge energy  HOG
  SVM   classification accuracy and speed
目   次
1 绪论    6
1.1 引言    6
1.2 选题的研究意义    6
1.3 SVM支持向量机    7
1.4 本文的主要内容和结构    8
2 知识背景    9 直方图相交内核支持向量机改进分类方法:http://www.lwfree.cn/jisuanjilunwen/20171113/15862.html
------分隔线----------------------------
推荐内容