摘要
为寻找更具鲁棒性和计算简便的特征描述子,提出了一种基于SIFT和MSE的局部聚集特征描述算法.分析说明了该方法在继承SIFT算法良好性质的基础上,通过对多尺度下信息熵的估计,能够快速准确找出图像局部结构特征并利用改进的非线性降维方法对特征描述子进行特征重划.实验结果表明,在图像尺度缩放、旋转、模糊、亮度变化等多种变换条件下,该描述子不仅能够取得更多的特征效果,并且计算速度较原算法大幅提升.该算法适用于实时性要求较高,存在旋转、尺度缩放、亮度差异等变换下的结构图像寻找描述子.
In order search a more robustness and convenient count method which shows a new feature descriptor algorithm is proposed in this paper. It analyzes and explains this way could quickly and accttrately to describe local slructure features based on inherit a higher quality of SIFF and MSE. In addition, it makes use of a changed locally linear embedding approach to process data so that it could reduce dimension. Experiment has drawn the conclusion that not only it could obtain more and betters the proposed descriptor but also the count speed could faster than SIFT for the image with zoom, rotation, blurring and illumination varying. This algorithm is suitable for searching the images which has structured features, when it exits multiple of varying.
出处
《电子学报》
EI
CAS
CSCD
北大核心
2014年第8期1619-1623,共5页
Acta Electronica Sinica
基金
国家自然科学基金(No.61175029
No.61203268
No.61202339)
关键词
多尺度熵
局部聚集特征
非线性降维
multiscale entropy
local aggregation features
locally linear embedding
