Abstract:

Aiming at the problem that the feature points extracted by traditional feature detection algorithms are poorly stable and not unique, resulting in excessive mismatch rate. Three criteria for selecting feature points are proposed：stability, uniqueness and significance criterion. The stability criterion guarantees to select robust and stable feature points. The uniqueness criterion rejects feature points with nearly duplicate features. The significance criterion preserves the points with salient feature information. By eliminating the feature points that do not meet the three criteria, it not only reduces the computation time of descriptors, but also improves the matching accuracy rate. Finally, the feature point selection criteria are applied to common matching algorithms. The experimental results show that the proposed feature point selecting method not only ensures the stability, uniqueness and significance of feature points, but also greatly improves the matching accuracy rate, repetition rate and matching speed.

Key words: image matching, feature points selection, stability scriterion, uniqueness criterion, significance criterion

摘要：

针对传统特征检测算法提取的特征点稳定性差且不唯一，导致误匹配率过高的问题，提出了三条特征点筛选准则：稳定性准则、唯一性准则和显著性准则。其中稳定性准则保证筛选出健壮稳定的特征点，唯一性准则剔除特征重复的特征点，显著性准则保留带有明显特征信息的点。通过剔除掉不满足这三条准则的特征点，不仅减少了描述符的计算时间，同时也提高了匹配正确率。将特征点筛选准则运用到常用的匹配算法中。实验结果表明，所提特征点筛选准则不仅保证了特征点的稳定性、唯一性和显著性，同时使得匹配正确率、重复率和匹配速度都有很大提高。

关键词: 图像匹配, 特征点筛选, 稳定性准则, 唯一性准则, 显著性准则