摘要
针对矩阵恢复用于红外偏振图像配准,速度较慢的劣势,提出了一种用于单幅图像快速配准的算法。选取上一次对相同背景图像配准后,得到的一列低秩向量为已知条件,将目标函数减少为求稀疏矩阵的1范数,通过迭代更新得到单幅相同场景下,目标存在一定变化或偏移的图像的配准参数,明显提高了配准的速度。实验结果表明,该算法配准变换参数误差小于0.02像素,单幅图像配准仅需3.31 s,且对噪声不敏感。
Aimed at the disadvantage that speed of infrared polarization image registration method based on matrix recovery is slow, a method used for fast registration of one image is put forward. A low-rank column which is acquired after last registration can be regarded as known condition and target function can be decreased to the 1 norm of sparse matrix. Registration parameter was achieved by one image whose target has changed or offset in the same scene after iterative updating. The speed of registration is highly increased. The result of experiment indicates that error of its registration parameters is less than 0.02 pixel. Running time of one image registration is 3.31 s, and the method is not sensitive to noise.
出处
《红外技术》
CSCD
北大核心
2014年第6期485-490,共6页
Infrared Technology
基金
国家自然科学基金
编号:41176158
关键词
图像配准
红外偏振
矩阵恢复
image registration
infrared polarization
matrix recover
