Abstract: Due to the invariance of scale, rotation and illumination, SIFT descriptor is widely used in computer vision, object recognition, medical image processing and other fields. But whatever methods taken, it is difficult to avoid mismatches. The second-order Gaussian algorithm is proposed to improve the accuracy of image matching after researching SIFT and RANSAC. Based on the Euclidean distance between feature points, the matching points are weighted by a Gaussian function. By comparing these weights, the points whose weight is not in the scale of a threshold are eliminated. The experimental results show that compared with RANSAC, the correct rate of the second-order Gaussian is significantly improved, and it has a good robustness to mismatches eliminated.

Key words: second-order Gaussian, image matching, mismatches eliminated

摘要： SIFT算子由于其良好的尺度、旋转、光照等不变性而被广泛应用于计算机视觉、目标识别、医学图像处理等领域。但无论采用何种算法，错配难以避免。针对错配点消除的问题，在对SIFT算法及RANSAC算法进行研究的基础上，提出二阶高斯算法。该算法根据特征点间的欧式距离，利用高斯函数为匹配点加权，筛除权值不在阈值范围内的匹配点。实验表明，与RANSAC算法相比，该算法的特征点正确匹配率明显提高，对错配点消除有较强的鲁棒性。

关键词: 二阶高斯, 图像匹配, 错配点消除