基于区域面积比的几何不变量构造算法  

Construction algorithm of geometric invariant based on area ratio

作　　者：邢昌元[1] 熊忠阳[2] 李月涵[3] 吴雪刚[1,4] 

机构地区：[1]长江师范学院计算机工程学院,重庆408100 [2]重庆大学计算机学院,重庆400044 [3]长江师范学院化学化工学院,重庆408100 [4]重庆大学通信工程学院,重庆400044

出　　处：《计算机应用研究》2017年第6期1900-1904,1908,共6页Application Research of Computers

基　　金：国家自然科学基金资助项目(61190122);重庆市教委科学技术研究项目(KJ15012028);长江师范学院校级科研项目(2015XJXM30)

摘　　要：几何不变量是物体几何结构信息的抽象与概括,在识别中可解决由目标移动产生的旋转或变形等问题。针对现有算法在提取几何不变量时,易造成误差增大和效率不高的情况,提出了一个基于区域面积比的几何不变量构造算法。该算法利用物体图像质心和凸补区域质心构成的直线,对灰度图像的面积区域应用划分策略构建几何不变量的矢量表示形式。在fish及coil-100数据集上的实验表明,算法得到的不变量特征满足仿射不变性,具有良好的区分辨别能力。Geometric invariant is the abstraction and generalization of geometrical information from the object, it solves the recognition problems generated by the target movement such as rotation or deformation. To solve the issue of error increase and low efficiency in feature extraction for geometric invariant algorithms, this paper proposed a construction algorithm of geometric invariant based on area ratio. First, the algorithm computed the coordinate positions of a centroid and a convex complement area＇ s centroid of a gray image after binaryzation. Then, it used a division strategy to partition the gray image to some areas according the line passes through these two centroids. Finally, it achieved the area ratio vector of geometric invariant by this algorithm. The experiments in the fish database and eoil-100 of Columbia University show that invariant extracted by the proposed algorithm satisfies affine invariance. Meanwhile, under a certain range of interference （ erase, smear and occlusion） , the invariant features extracted from the object image have discrimination ability.

关 键 词：模式识别 特征提取 几何不变量 凸包 凸补区域 

分 类 号：TP391.7[自动化与计算机技术—计算机应用技术]