Abstract: In order to obtain high accurate，high stable registration core image，a registration method based on the edge characters and Keren algorithm is proposed.Firstly，the edges of reference images and images requiring registration are extracted.Then，these images are decomposed into Gauss pyramid structure.Finaly，the Keren algorithm is applied to core image registration from coarse to fine accuracy.The experiment shows that the method compared with Keren algorithm，they both have high registration accuracy in the case of a small rotation angle，but the former has lower absolute error in the case of a large rotation angle，and makes the absolute error of translation parameters under 0.1 pixel and under 0.1 degree in the case of a large rotation angle.So，the proposed method improves the accuracy of registration of the core images.

Key words: image registration, Sobel operator, Keren improvement algorithm

摘要： 为获得高精确度、高稳定度的岩心配准图像，提出利用Sobel算子提取图像边缘并结合keren算法的图像配准方法。该方法首先提取参考图像和待配准图像边缘，接着对边缘图像进行高斯金字塔分解，最后，利用keren改进算法实现由粗到细的岩心图像配准。实验结果表明该算法与Keren算法相比，在小角度下都有较高的配准精度，大角度旋转情况下该方法绝对误差明显降低，且在大角度旋转下平移参数和角度参数分别可获得0.1个像素以下和0.1度以下的绝对误差精度，提高了岩心配准精度。

关键词: 图像配准, Sobel算子, Keren改进算法