针对复杂场景中目标由于成像畸变、部分遮挡或局部缺失难于识别的问题,提出一种新的仿射配准算法。首先给出了CCCTI码(cyclic code of corner and tangent and inflexion points)的定义,该码易于确定模型和目标轮廓上关键特征点的对应关系;其次利用关键特征点对轮廓进行分段,根据对应子曲线段的两端点及其形心估算变换矩阵,再引入谱系聚类法对所有估算矩阵进行聚类,降低最终估算矩阵的误差,且使算法适用于部分遮挡或局部缺失,提高算法的鲁棒性;最后计算能够聚类的对应子曲线段的总形心,并利用总形心与对应子曲线段的两端点再次估算变换矩阵,提高配准的精度。理论分析和实验结果均表明,该算法能有效地进行仿射配准,并能处理部分遮挡或局部缺失。

It is difficult to recognize objects when they are distorted and partially occluded or broken. In order to solve this problem, a new registration algorithm is proposed in this paper. First, we define the CCCTI(cyclic code of corner and tangent and inflexion points) of a planar curve. This code can easily determine the corresponding relation of the key feature points between the model and target contour. Second, contours are segmented by using the key feature points, and the transformation matrix is estimated based on two endpoints of the sub-curve and its centroid. Then, all estimate matrices are clustered using hierarchical clustering method. This process can reduce the final error of the estimation matrix and improve the robustness of the algorithm, making the algorithm suitable for partial occlusions or broken objects. Finally, we calculate the total centroid of corresponding sub-curve segments, and estimate the transformation matrix again using the total centroid and two endpoints of the corresponding sub-curve segments, to improve the accuracy of the registration. The theoretical analysis and experimental results show that our algorithm can be effectively used for affine registration, and that it can deal with partial occlusions and broken objects.