Point pattern matching algorithm based on local topological characteristic and probabilistic relaxation labeling

To reduce the impact of outliers and noises on point pattern matching, a novel point pattern matching algorithm based on local topological characteristic and probabilistic relaxation labeling (LTC-PRL) is proposed in this paper. For each point in a point set, partial adjacent points are used to describe its local topological characteristic. To avoid the defects in angle coding of the existing global topological characteristic, a binary adjacent code is adopted in the local topological characteristic. And since the assignment of angle is greater than the distance, bigger weight is given to the angle while computing the similarity of the local topological characteristic among points. Finally, a robust compatibility measurement is defined and the support function is iterated by probabilistic relaxation labeling to get the best matching result. Experiments on synthetic data and the real image data show that the LTC-PRL has great matching performance when outliers and noises exist.