摘要
为了在一定亚像元匹配精度下提高亚像元匹配速度、研究传统的模板匹配亚像元方法,提出将图像重采样法和曲面拟合法结合起来用于亚像元匹配。先利用序贯相似性检测法,找到像元级的最佳匹配点,然后对模板n倍重采样,计算每个子模板与最佳匹配点的归一化相似性度量值,并计算当归一化相似性度量值最大时所对应的子模板在图像中的位置,再在绝对误差值最小的3×3区域内应用曲面拟合法,从而得到亚像元定位坐标。结果表明,采用图像重采样法和曲面拟合法相结合的新方法,在10倍重采样的前提下,匹配精度高于0.04像元的亚像元精度。
To increase sub-pixel matching speed with a certain sub-pixel accuracy, the conventional algorithms to obtain sub-pixel accuracy in template matching was discussed, a new algorithm was proposed to combine the resampling and surface fitting methods. At first, it should use the sequential similarity detection algorithm to find the best match point, then only the model was need to be resampled n-times,next the normalization correlations among each sub-template and the image in the best match point was calculated, and the position of the sub-template which was corresponded to the maximum normalization correlation in the image was calculated. At last the method of curve fitting in the 3 × 3 area of that position was applied to obtain the absolute coordinates of the positioning with sub-pixel accuracy. The experiment shows that the accuracy obtained by the new method base on 10 times resampling is less than 0.04 pixcl.
出处
《激光技术》
CAS
CSCD
北大核心
2008年第5期493-495,共3页
Laser Technology
关键词
信息光学
亚像元
重采样
遥感图像
曲面拟合
information optics
sub-pixel
resampling
remote sensing image
curve fitting
