Li Bin, Wang Wei, Ye Hao
Department of Automation, Tsinghua University, Beijing, 100084, China.
Opt Express. 2013 Apr 22;21(8):9824-38. doi: 10.1364/OE.21.009824.
A new automatic feature-based registration algorithm is presented for multi-sensor images with projective deformation. Contours are firstly extracted from both reference and sensed images as basic features in the proposed method. Since it is difficult to design a projective-invariant descriptor from the contour information directly, a new feature named Five Sequential Corners (FSC) is constructed based on the corners detected from the extracted contours. By introducing algebraic projective invariants, we design a descriptor for each FSC that is ensured to be robust against projective deformation. Further, no gray scale related information is required in calculating the descriptor, thus it is also robust against the gray scale discrepancy between the multi-sensor image pairs. Experimental results utilizing real image pairs are presented to show the merits of the proposed registration method.
针对具有投影变形的多传感器图像，提出了一种基于特征的自动配准新算法。在所提方法中，首先从参考图像和传感图像中提取轮廓作为基本特征。由于直接从轮廓信息设计投影不变描述符较为困难，因此基于从提取轮廓中检测到的角点构造了一种名为“五个连续角点（FSC）”的新特征。通过引入代数投影不变量，我们为每个FSC设计了一个描述符，确保其对投影变形具有鲁棒性。此外，在计算描述符时不需要与灰度相关的信息，因此它对多传感器图像对之间的灰度差异也具有鲁棒性。给出了利用真实图像对的实验结果，以展示所提配准方法的优点。
