摘要
为了在计算机视觉任务中构造有意义的图像表示,提出一种基于概率密度函数(p.d.f)梯度方向直方图特征的分层稀疏表示方法用于图像分类。传统分层稀疏表示方法利用SIFT描述子或者直接从图像块学习图像表示,通常不具有较强判别性。该文利用具有通用性的p.d.f特征进行分层学习并使用空间金字塔最大池化方式构造图像级稀疏表示。实验结果证明了所提算法的鲁棒性和有效性,在UIUC-Sports,Oxford Flowers,Scene15三类数据集上分别达到87.3%,86.6%,84.1%的分类准确率。
In order to construct the meaningful image representation in computer vision task,a novel hierarchical sparserepresentation method based on oriented histogram feature of probability density function(p.d.f)gradients is proposed for imageclassification.The traditional hierarchical sparse representation method which learns the image representation with SIFT descrip-tor or learn it directly from image block has poor discrimination.A universal p.d.f feature is employed for hierarchical learning,and the spatial pyramid max pooling method is used to construct the image-level sparse representation.The experimental resultsshow that the algorithm has robustness and availability,and the classification accuracy for classifying the datasets of UIUC-Sports,Oxford Flowers and Scene15can reach up to87.3%,86.6%and84.1%respectively.
作者
王博
WANG Bo(School of Physics and Electronic.Electrical Engineering,Ningxia University,Yinchuan 750021,China)
出处
《现代电子技术》
北大核心
2017年第10期95-98,102,共5页
Modern Electronics Technique
基金
国家重点基础研究发展计划(国家"973")项目:网络大数据感知融合与表示方法研究(2014CB340403)
关键词
图像分类
分层稀疏表示
空间金字塔最大池化
图像表示
image classification
hierarchical sparse representation
spatial pyramid max pooling
image representation
