基于局部统计特性及SIFT算法的图像配准

摘要

In order to overcome the great computational complexity in SIFT (i.e.,scale invar-iant feature transform) image matching algorithm,the paper has been designed to firstly estimate the local information function of two images to be matched,and subsequently to apply SIFT fea-ture exaction in the areas of larger amount of local information whereas not to utilize it in the are-as of small amount of local information.A relational formula for geometric calibration has been established by using the correspondence between detected feature points with the aim of matching their images.The theoretical analysis and experimental results show that most SIFT feature points can be obtained through detecting SIFT features in less image positions by using local in-formation functions to prescreen them.This method can further improve the speed of computa-tions through giving up some feature points with lesser information.This algorithm achieves good matching effect.%针对SIFT算法运算量较大的问题,提出了一种改进算法,可以减少基于SIFT算法的图像匹配的运算量.首先定义图像的局部信息函数,使得局部信息函数数值大小反映该点是特征的概率的大小.利用一个阈值将局部信息函数进行分割,进而在局部信息较大的区域进行SIFT特征提取,局部信息小的区域不进行SIFT特征提取,从而达到提高SIFT特征点的检测效率.利用检测到的特征点的对应关系建立几何校正关系式,进而完成图像匹配.理论分析及实验表明,通过局部信息函数的预筛选功能,只用在较少的图像位置进行SIFT特征检测,就能得到绝大部分SIFT特征点.并能通过舍弃一些信息量较少的特征点,换取速度的进一步提升.算法达到了预期的效果.