摘要
针对光学与合成孔径雷达(SAR)图像难以配准的问题,提出一种基于空间约束的尺度不变牲变换(SIFT)算法。该算法对光学图像和SAR图像分别进行预处理,包括利用增强Frost滤波抑制SAR图像的相干斑噪声,及运用自适应直方图均衡法增强光学和SAR图像之间的共性轮廓特征。人工选取3个~4个同名控制点对进行粗配准。通过改进的SIFT方法提取特征点,以结构相似性指数作为特征点之间的相似性测度,并采用kd-tree搜索策略得到初始匹配点对。使用空间约束条件和随机抽样一致性算法筛选匹配点对,利用最终的精匹配点对完成配准。实验结果表明,该算法对光学和SAR图像的配准可以取得较高的精度,配准精度优于2个像素。
Aiming at the problem that optical and Synthetic Aperture Radar(SAR)images is difficult for registration,this paper proposes an improved Scale Invariant Feature Transform(SIFT)algorithm based on spatial constraints.The algorithm preprocesses optical and SAR images,the enhanced Frost filter and adaptive local histogram equalization are adopted to improve the common property of the optical and SAR images,and three or four corresponding points are selected manually to realize a coarse registration.An improved version of the SIFT method is proposed to detect the key points,and the Structure Similarity(SSIM)is used to obtain initial matching features by kd-tree search method.The initial matching features are refined by spatial constraints and Random Sample Consensus(RANSAC)method.The refined feature matches are used for registration.Experimental results show that the proposed method has higher precision for optical and SAR images registration and registration precision achieves 2 pixels.
出处
《计算机工程》
CAS
CSCD
北大核心
2015年第12期182-187,共6页
Computer Engineering
基金
国家自然科学基金资助项目(41301383)
关键词
尺度不变特征转换
空间约束
结构相似性指数
合成孔径雷达
图像配准
Scale Invariant Feature Transform(SIFT)
spatial constraints
Structure Similarity(SSIM)index
Synthetic Aperture Radar(SAR)
image registration