图像特征提取中领域尺寸和本征维数的自动选择算法  

Auto-selected algorithm of the neighborhood size and intrinsic dimension for image feature extraction

作　　者：刘凤连[1] 汪日伟[1] 程俊[1] 康绢鹏 

机构地区：[1]天津理工大学计算机视觉与系统教育部重点实验室和天津市智能计算及软件新技术重点实验室,天津300384

出　　处：《光电子．激光》2013年第12期2416-2420,共5页Journal of Optoelectronics·Laser

基　　金：国家自然科学基金(60872064;61102125);天津市自然科学基金(12JCYBJC10200);大学生创新训练计划项目(201310060013)资助项目

摘　　要：本文首先对图像特征提取过程中的近邻距离进行线性重构,然后对其在整个流形上的分布进行优化运算,得到一个最优线性重构权为变量的图像数据局部低维特征的表达函数。最后以该函数稳定性的最小度量构造出适用于图像特征提取的邻域尺寸和本征维数的自动选择策略。实验表明该方法实现了图像数据本征维数与邻域尺寸的自动化选择,并具有计算简单、匹配率高且计算复杂性低等特点。Dimensionality reduction algorithms have been applied widely in computer vision,medical image processing, video processing, face recognition, image retrieval etc. However, some problems need to be fixed. First of all,how to get the domain size and intrinsic dimension is first primary problem to be fixed, which is seriously restricted the rapid development. Traditonal method used the K-Nearest Neighbor to search the neighborhoods of each image sample. But it costs much time sometimes. In this paper, we aim at the problem of finding intrinsic structure and domain size automatically for high dimensional image data. We present a new technique which can get intrinsic dimension and neighborhood size automatically and adaptively. The algorithm can be used for extracting local features from images. Firstly, we made lin ear reconstruction based on the nearest neighbor distance for image feature extraction,and optimize the distribution on whole manifold, then we get expression function in which variable is the optimal linear reconstruction for locally low dimensional feature. Lastly,minimiing the variance of the function is to get aotomatie selection strategy. Experiments show that the algorithm is not only simple but also high matc hing rate and low computational complexity.

关 键 词：本征维数 邻域尺寸 图像特征 流形学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]