摘要
针对电子稳像实时性和鲁棒性的要求,提出了一种快速的单象限Fourier-Mellin域相位相关的近似运动估计算法,即首先根据Fourier变换振幅谱的对称性和空间域图像块匹配的思想,提出了一种新颖的单象限Fourier-Mellin变换方法,即在频率域中采用单象限振幅谱进行对数极坐标变换;然后结合相位相关技术估算帧间运动参数。与基于标准Fourier-Mellin变换的算法相比,在保证精度的前提下,新算法可大幅度降低算法的计算复杂度。实验结果表明,该新算法的运行速度比标准算法提高了约75%,且精度略有提高。
To realize fast and robust electronic image stabilization, an algorithm using phase correlation in single-quadrant Fourier-Mellin domain to estimate similarity motion parameters is presented. First, according to the property of origin symmetry of the Fourier magnitude spectrum and the idea of image block matching in spatial domain, a novel method of single-quadrant Fourier-Mellin transformation （SQFMT） is proposed, which applies log-polar transformation in frequency domain to map the single-quadrant magnitude spectra from Cartesian coordinate space to log-polar one. The similarity transformation parameters are estimated using SQFMT combined with phase correlation method. In contrast to the approach based on the standard Fourier-Mellin transformation （FMT） , our algorithm yields an effective solution in computational performance, and achieving consistent accuracy. The experimental results show that our algorithm can achieve a reduction of 75% run time on average, with a small accuracy increases comparied with standard method.
出处
《中国图象图形学报》
CSCD
北大核心
2009年第9期1790-1795,共6页
Journal of Image and Graphics
基金
辽宁省自然科学基金项目(20062191)
浙江大学CAD&CG国家重点实验室开放基金项目(A0906)
