利用仿射几何的仿射不变特征提取方法

摘要

提出一种新的基于仿射几何的仿射不变特征提取方法,适合于目标的识别和图像的匹配.算法分3步执行:首先,提取区域的质心和扩展质心,质心和扩展质心的连线把目标区域分割成两部分,再分别计算两部分区域的质心,如此迭代,直到提取出满足要求的质心个数;然后,依次计算四边形的面积,其中四边形的顶点分别为其连线分割目标区域的两个质心和两个分割区域的质心;最后,依次计算各个四边形的面积比,得到仿射不变特征矢量;另外,仿射变换的参数也可以通过计算提取的质心坐标得到.实验表明,提取的不变特征矢量稳健性好、计算速度快、分类精度高.%A new affine invariants based on affine geometry for object recognition and matching is proposed in the paper.The algorithm can be divided into three following steps. Firstly, extract the centroid and extend centroid of the object region. The connecting line between them thus devides the region into two parts, and the centroids of the two parts will be recalculated. Iteratively, we can obtain the demand centroids that satisfy the requirement. Secondly, calculate the areas of the quadrangles, whose vertexes are the centroids of the regions cut by the connecting line and the centroids of the two parts. Thirdly, calculated the ratios of the areas and the affine invariant vector is acquired Besides, the parameters of the affine transform can also be computed by the coordinates of the centroids. It is proved by the experiment that the extracted affine invariants are of robustness, efficiency and high classification accuracy.