与光学图像相比,SAR图像噪声干扰明显增强,导致光学图像处理领域中常用的尺度不变特征提取算法的稳定性在SAR图像中明显下降。为此,提出一种基于Harris算子的SAR图像尺度不变特征提取方法。该方法首先将单尺度图像特征点检测中具有良好稳定性的Harris算子拓展到多尺度图像域,并结合Harris算子极值点的稳定性分析实现尺度不变特征定位,以提高特征定位的稳定性。进而采用迭代滤波器取代传统卷积滤波器完成特征定位中所需高斯滤波操作,以提高特征定位的速度。最后利用特征点邻域内的像素梯度信息完成了特征描述字的构造。SAR图像实测数据验证表明,与其他尺度不变特征提取算法相比,本文算法具有更优的稳定性。

Because of the stronger noise, the stable performance of optical method degrades obviously in SAR image scale invariant feature extraction. To deal with it, we propose a new SAR image scale invariant feature extraction algorithm based on Harris operator. As well known, Harris operator has well stable performance in single scale image feature point detection. In this paper, we first adapt it to multi-scale image and select the features positions by analyzing the stability of Harris measure local maximum points. Then to improve the speed,the recursive Gaussian filter is designed to replace the convolution filter. At last, the feature descriptor is designed based on the gradient information of feature point neighborhood. Based on real SAR images, the experimental results show that an obviously stable performance improvement can be obtained through our algorithm.