基于特征匹配的鲁棒图像镶嵌

Feature-Based Robust Image Mosaicing

提出了一种联合相位匹配和特征匹配的鲁棒性配准算法.在利用相位相关算法获得两幅图像的大致重合位置以后,通过Harris特征跟踪算子获得大量的对应特征点,然后利用最小中值二乘法来剔除前面匹配过程中的局外点.理论分析和实验对比表明,该算法能够在复杂场景的精确匹配中获得较好的效果,在速度和精度之间取得良好的平衡.

The paper proposed a new robust registration algorithm that cooperates both phased-based and feature-based algorithm. After getting an approximate overlap area of image pairs, Harris tracer is used to get lots of corresponding features. Next, least median square method is employed to reject the outliers that inevitably incurred in the previous steps. Contrast of the theory and experiments shows that the proposed algorithm can get good effects in complicated scenes and high precision matching, and has a good tradeoff between speed and precision.