基于多尺度角点的改进SIFT算法

AN IMPROVED SIFT ALGORITHM BASED ON MULTI-SCALE CORNER POINT

重点讨论一种基于角点的改进SIFT(Scale Invariant Feature Transform,即尺度不变特征变换)算法。该算法采用统一的低主曲率比值删除不稳定边缘响应点,把高斯空间中提取的角点加入到运用主曲率比值筛选后的SIFT特征点中。另外,在角点检测中,以图像区域方差来动态确定角点检测的阈值,大大提高了算法的适应性。实验证明,改进后的算法能提取更加稠密且高匹配的特征点,并且具有对主曲率比值不敏感的优点。

This paper focuses on a kind of improved SIFT algorithm based on the corner point. The algorithm deletes the unstable edge response points by using the unified low main curvature ratio, and combine the SIFT features points which have been screened according to main curvature ratio with the corner points of the image space of Gaussian. In addition, on the aspect of the corner detection, the variance of image region dynamically determines the threshold of the corner detection, so that greatly improving the adaptability of the algorithm. Experiments show that the improved algorithm can extract more dense and high matching feature points, and has the advantage of being inserisitive to the main curvature ratio.