基于区域边缘统计的图像特征描述新方法  

A New Image Feature Descriptor Based on Region Edge Statistical

作　　者：余旺盛[1] 田孝华[1] 侯志强[1] 

机构地区：[1]空军工程大学信息与导航学院,西安710077

出　　处：《计算机学报》2014年第6期1398-1410,共13页Chinese Journal of Computers

基　　金：国家自然科学基金(61175029);陕西省自然科学基金(2011JM8015)资助~~

摘　　要：直方图特征是计算机视觉领域最为常用的图像特征之一,然而空间统计信息的缺乏导致传统灰度直方图对图像特征的区分能力较差.对此,文中通过引入直方图子区间对应区域的边缘复杂度概念,提出了一种描述图像特征的区域边缘直方图方法.文章首先定义了区域边缘复杂度的概念,并提出以直方图各子区间的边缘复杂度作为空间权系数对各子区间进行加权,然后通过对加权后的直方图进行化简进而得到区域边缘直方图的简洁描述形式,最后,给出了基于积分直方图实现的快速计算方法.实验结果表明,区域边缘直方图方法对图像特征的区分能力明显优于灰度直方图,克服了空间直方图等改进方法对图像旋转的敏感性,运算效率比空间直方图提高了两个数量级以上.Histogram is one of the most popular image features which are widely used in computer vision.However,lacking of space information in traditional gray histogram leads to an unsatis factory performance in discriminating image features.To resolve this problem,this paper intro duces the region edge complexity of histogram ＇ subinterval and proposes a new method to describe the image features,namely,region edge histogram.Firstly,it defines a new concept of region edge complexity,and weighs each of the histogram bins using its corresponding region edge complexity.Then,it simplifies the expression of the weighted histogram and derives a compact form.Lastly,it presents an accelerated method to compute the region edge histogram based on integral histogram.The experimental results indicate that region edge histogram is superior to gray histogram and its ameliorations in feature discrimination and robust to the image rotation.Besides,it improves the efficiency at least two orders of magnitude than spatial histogram.

关 键 词：模板检测 视觉跟踪 空间直方图 区域边缘直方图 边缘复杂度 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]