SIFT在高分辨率SAR图像自动配准中的性能分析

Performance analysis of SIFT to registration of high-resolution accuracy SAR imagery

合成孔径雷达(SAR)图像的自动配准长期以来都未能很好的解决,特别是高分辨率SAR图像其配准的关键是稳健的特征提取与特征匹配算法。在光学图像配准中,最常用的特征点提取算法是Harris算子,而近年来SIFT(尺度不变特性变换)算法也因其优越的性能成为当前比较流行的算法。探讨了Harris和SIFT特征提取算法在高分辨SAR图像自动配准中的应用,并选取4对有代表性的SAR图像进行了配准实验,对2种特征提取算法的运行时间、所提取匹配点对的正确率以及特征点的提取精度进行了比较。通过定性及定量分析,在同轨获取的高分辨率SAR图像配准中,SIFT均能实现精确配准,其适用性及精度均优于Harris。

Automatic registration of SAR especially high-resolution SAR imagery has not been well solved till now, the key to automatic registration is extracting stable feature and matching feature. Harris operator is the most common feature extraction algorithm in the optical imagery registration, and SIFT（Scale Invariant Feature Transform）algorithm become more popular for its superior performance in recent years. In this paper, SIFT algorithm aiming to the SAR imagery＇s characteristic is compared with traditional Harris comer point achieve arithmetic in four representative designed experiments. Two algorithms were compared in running time,the correct rate of the matching and the accuracy of the key points. According to qualitative and quantitative analysis, SIFt can accurately register the high resolution SAR images were acquired in the same ascending or descending orbit, and it is better than Harris in applicability and precision.