改进的SIFT特征提取和匹配算法

摘要

A robust automated image matching strategy based on an improved SIFT feature extraction algorithm was proposed according to the characteristics of SIFT algorithm and lunar images. Firstly, the extraction equilibrium of key points was improved by automatically adjusting the coefficient of controlling contrast in the SIFT algorithm. Then, the coarse matching was carried out by using the ratio of the minimum and the second minimum Euclidean distance between the description vectors of SIFT as a threshold and the part incorrect matching of the coarse matching set was removed by principal direction angle difference histogram of matches. Finally, the initial parameters of perspective transformation model were determined by using modified RANSAC method, automatically calculated random sampling numbers, and parameters of error tolerance. Moreover, the model and the error tolerance were used to calibrate the matching pairs in the matching set and to select the correct matching pairs. The experimental results prove that the proposed method is stable and reliable under some variationsfor view points, illumination, rotation, scale and out of focus and it can select the matching threshold of images automatically without manual intervention.%针对月球影像和尺度不变特征变换(SIFT)算法的特点,在改进SIFT特征提取算法的基础上,提出了一种稳健的图像自动匹配策略.首先,自动调整SIFT算法中的对比度控制系数,提高关键点提取的均衡性；然后,用SIFT描述向量之间的欧氏距离最小值与次小值的比值作为阈值,进行粗匹配,并利用匹配对主方向角度差直方图剔除部分伪匹配；最后,自动计算随机采样次数、误差容忍度等参数,使用改进的随机选取一致性(RANSAC)方法确定透视变换模型初始参数,并用该模型和误差容忍度对匹配集合中的匹配对进行校验,选取出正确的匹配对.实验结果表明,在图像有一定程度的视点、光照、旋转、比例、模糊和对比度变化等情形下,该算法稳定、可靠.该方法能有效解决图像匹配门限的选择问题,真正实现了无人工干预的自动匹配.