摘要
直接使用傅里叶梅林变换计算出的图像间旋转参数的精度不高,以此旋转参数对图像进行纠正后再使用相位相关的扩展方法计算出的图像间的平移参数精度也不高.针对这种情况,提出一种改进的角度检测方法.传统的方法是先对图像进行傅里叶变换,然后进行极坐标变换,再使用相位相关得到相关谱,通过相关谱的峰值位置确定图像间的旋转角度,最后对图像做角度纠正并计算平移参数.而本文对相关谱峰值所在区域进行拟合,使用拟合后的峰值位置替代原来的峰值位置确定图像间的旋转角度.由于拟合后的峰值位置的精度高于拟合前,估计角度的精度会更高,使用此角度做旋转纠正能够提高图像之间的相关性,进而也提高平移参数的精度.实验证明:与Fourier-Mellin变换的通常做法相比,此方法显著提高了配准的精度,平均角度精度提高0.187°,水平方向上平均平移精度提高0.339 pixel,垂直方向上平均平移精度提高0.296 pixel.
The accuracy of the rotation angle parameter calculated directly by using Fourier-Mellin transform is not high. And the trans- lation parameters calculated by phase correlation or extension of phase correlation between images which has corrected by using the an- gle from direct use of Fourier-Mellin is also not high. In view of this situation, a more precise angle detection method is proposed. The process of traditional method is as follows： at first, using Fourier transform to convert the image to the frequently domain, converting the magnitude spectrum to polar coordinates, using F~ and phase correlation to calculate the correlation spectrum, getting the rotation angle by using the peak position of correlation spectrum, and then using this angle to correct images, finally calculating translation pa- rameters. This paper has used some methods to fit the peak area, and find the more precise location of the peak, and use it to calculate the angle other than the original peak position, and the accuracy of the rotation angle is improved significantly. Experimental results show that the proposed method can not only get more precise rotation parameters, but also more precise translation parameters, and the accuracy of registration is improved greatly. The average rotation angle error is reduced by 0.17o, and the horizontal translation error and the vertical error is decreased by, respectively ,0.339 pixels and 0. 296 pixels.
出处
《小型微型计算机系统》
CSCD
北大核心
2015年第12期2763-2768,共6页
Journal of Chinese Computer Systems
基金
国家科技支撑计划项目(2012BAH31B01)资助
北京市自然科学基金重点项目(B类)(KZ201310028035)资助
关键词
亚像素配准
傅里叶梅林变换
曲线拟合
极坐标变换
Sub-pixel registration
Fourier-Mellin transform
curve fitting
polar coordinate transformation
作者简介
E—mail：lihanlun@126．com,作者简介：李含伦，男，1988年生，博士研究生，研究方向为遥感方向
张爱武,女，1972年生，教授，博士生导师，研究方向为空间信息获取与处理、三维激光成像、高光谱成像等
孟宪刚，男，1984年生，博士研究生，研究方向为遥感传感器系统集成
胡少兴，男，1972年生，博士研究生，副教授，研究方向为传感器集成、三维激光成像等．
