基于插值和曲面拟合的图像亚像素配准算法

An Sub-pixel Precision Registration Algorithm Based on Image Interpolation and Fitting

图像配准在很多领域有着广泛的应用,如在计算机视觉、识别模式、军事、医学成像方面等方面,而且像素级的配准精度远远不能满足要求,需要达到亚像素级的图像配准。本文在传统三次曲面拟合的基础上,提出了一种改进的亚像素精度配准算法,该算法首先找到像素级配准位置(u_1,v_1),之后在附近区域进行灰度插值,在图像插值过程中采用的是具有在节点处具有二阶连续导数的三次样条插值和双立方插值法。最后经过曲面拟合求得亚像素配准结果。实验结果表明,与传统的曲面拟合算法相比,本文提出的算法在精度上得到了明显的改善,精度可以达到0.01pixel。

Image registration is applied extensively in many fields, such as in computer vision, recognize patterns, military, medical imaging, and so on .registration at pixel level can＇t meet our requirements, registration Should be at Sub-pixel level. An improved sub-pixel registration algorithm is proposed based on cubic surface fitting method. First, find the pixel registration coordinates （v1, v1）. Second, there is an image interpolation process. We use cubic spline interpolation and bicubic interpolation in the image interpolation process, and cubic spline interpolation has the second-order continual derivative in the node. Finally, we get the sub-pixel registration results rely on cubic surface fitting method. The experiment results show that the algorithm relatively to the general cubic surface fitting has a slight increase in calculation error. The experimental results show that the algorithm maintains a high precision of sub-pixel extraction.