【摘要】 文章提出了一种超光谱图像的高精度自动配准算法。该算法采用多元互信息作为相似性测度,能同时利用多个波段图像的可知信息,并能很好地克服图像光谱特征变化的影响;同时,随机优化算法二阶同步试探随机逼近算法(2SPSA)的应用解决了多元互信息的多参数优化问题;另外,算法的多分辨率实现形式,能明显加快搜索速度并增强优化算法的鲁棒性。实验结果表明该算法能有效地处理超光谱图像配准问题,并能达到亚像素的配准精度。更多还原

【Abstract】 A new registration algorithm for hyperspectral imagery is proposed,which is especially targeted for automated implementary fashion and high accuracy.Firstly,we employ multivariate mutual information as the similarity metric so that the registration algorithm can make efficient use of the available information within several band images and can be robust against effects caused by differences of spectral characteristic in band images.Then the multi-parameter optimization problem for multivariate mutual information is solved using a stochastic optimization technique called 2SPSA(Second-order Simultaneous Perturbation Stochastic Approximation).A multiresolution implementation of the registration algorithm accelerates search speed and improves robustness of our optimization technique significantly.The experimental results show that this algorithm is efficient for registering hyperspectral imagery and can yield sub-pixel accuracy.更多还原