摘要
针对现有频域法FFT和PPFFT对于大倍率图像配准精度和计算效率难以同时保证的问题,研究了一种基于多重频谱叠加的频域算法,为最大限度减小插值误差噪声,该算法采用多分辨率频谱插值,构造出更接近对数极坐标网格。文中给出分辨率参数设置和算法流程。最后通过算例验证,结果表明:本算法对无裁减图像最大配准倍率可至10,对裁减图像可至5,而FFT和PPFFT(无迭代)对于两种类型图像都约为2;相对于特征点匹配法,本算法也适用于更大倍率的图像配准,从而验证了该算法的准确性和可行性。
Of frequency-domain method, registration accuracy and efficiency of FFT or PPFFT for large scale image is difficult to be obtained simultaneously. With this problem considered, a new registration method based on overlaying of multiple spectrums is proposed, which interpolates spectrums of multi resolution and generates the grid closest to log-polar, to minimize interpolation error noise. In addition, how to set resolution parameter and the algorithm flow is also explained. Finally, to validate the accuracy and feasibility of the algorithm, three examples were calculated through the new algorithm, and the result shows as followed： as for the image without cutting, the maximum registration scale of the improved method without iterations can reach 10, and as for the image with cutting, the same one can reach 5, while the FFT and PPFFT （non iterative） for the two kinds of image is about 2; compared with the feature point matching method, the improved method is also applied to image with larger scale.
出处
《红外技术》
CSCD
北大核心
2015年第5期424-430,共7页
Infrared Technology
基金
中国博士后科学基金
编号:2012T50870
关键词
多重频谱
大倍率配准
对数极坐标
插值误差
multiple spectrum, large scaleregistration, log-polar ordinary, interpolation error
