摘要
针对词袋模型统计聚集算法忽略了编码矢量的其它统计特征信息及空间信息,并且只能与常用核函数相配合度量图像之间相似性的问题,该文提出一种基于空间概率乘积核函数的图像分类(SPPKBIG)算法。使用Parzen窗方法估计编码矢量所服从的概率密度分布,用来描述图像内容,使用空间概率乘积核函数构建图像之间的核矩阵,最后使用基于此核矩阵的支持向量机对图像进行分类。实验结果表明,SPPKBIC算法对15类场景数据集和MSRcv2数据集的平均分类正确率分别为84.1%和94.8%。
Aiming at the problems that the statistic pooling method using bag-of-words （ BoW） discards a lot of statistical and spatial information of coded vectors and only interacts with the standard kernel function to measure similarities of images,a spatial probability product kernel based image classification（ SPPKBIC） algorithm is proposed here. The probability distributions of coded vectors are estimated by Parzen window method to describe images. The kernel matrices of images are calculated using the spatial probability product kernel function. Images are classified by support vector machines based on the kernel matrices. The experimental results show that the average classification accuracy of the SPPKBIG algorithm for scene 15 dataset and MSRcv2 dataset reach 84. 1% and 94. 8% respectively.
出处
《南京理工大学学报》
EI
CAS
CSCD
北大核心
2014年第3期325-331,共7页
Journal of Nanjing University of Science and Technology
基金
国家自然科学基金重大研究计划(9082030)
国家自然科学基金青年项目(61103059)
关键词
空间概率乘积核函数
图像分类
词袋
统计聚集算法
统计特征信息
空间信息
Parzen窗方法
概率密度分布
核矩阵
支持向量机
spatial probability product kernel
image classification
bag-of-words
statistic pooling method
statistical information
spatial information
Parzen window method
probability distributions
kernel matrices
support vector machines
