SIFT特征匹配算法改进研究

为了适应于景象匹配导航及制导等实时性要求较高的领域,对SIFT特征匹配算法进行改进,提出了基于D^2OG特征点检测算子的改进的SIFT特征匹配算法。改进算法用D^2OG金字塔的过零点检测代替DOG金字塔的极值点检测提取尺度不变特征点,巧妙简化高斯金字塔的结构,降低了算法复杂度和时间代价。以标准测试图库中大量不同几何和灰度畸变图像为基础的仿真实验表明,基于D^2OG特征点检测算子的改进的SIFT特征匹配算法在保持原算法鲁棒性和精度的前提下,较大的提高了算法实时性。

Abstract：

An improved Scale Invariant Feature Transform algorithm was proposed based on D^2OG interest point detector for better real time performance in the application of scene matching navigation and so on. In order to detect the scale invariant interest point, a D^2OG pyramid is built and extreme detection in the DOG pyramid was replaced by zero detection in the D^2OG pyramid, which simplified the structure of DOG pyramid, so as to lower the complexity of algorithm, lessen the running time. Numerous experiments were carried out on standard testing images under various shooting conditions such as geometric distortion, illumination variation and so on. The result shows that the method has a big progress in the real time performance compared to the original one, with equally robustness and precision.