带标志位的均匀性特征检测SIFT配准算法研究

Research of well distributed feature detection with label of scale-invariant feature transform algorithm

针对尺度不变特征SIFT配准算法中检测到的特征点不具有均匀分布的特性,实现了均匀性特征检测方法,同时对像素点设置标志位对检测步长进行动态调整。均匀性特征检测方法能够检测到更有效、更具有代表性的特征点,从而得到更加精确的图像变换关系;设置标志位对动态步长进行调整,可以进一步减少检测的次数。将带标志位的均匀性特征检测SIFT算法应用于图像的配准,实验表明改进算法的性能得到了有效提高。

The method of well-distributed feature detection is taken in view of the situation in which the features detected in the scale invariant feature transform algorithm of image registration are not well distributed, and the dynamic step for each pixel is regulated by setting labels at the same time. More efficient and representative features can be detected by taking the method of well distributed feature detection, so the transformation relationship between the images can be more accurate; Setting labels to adjust dynamic step can reduce the times of feature detection. Experiments indicate that the performance of the improved algorithm has been efficiently improved by applying the well distributed feature into detection with label image registration.