SAR与全色图像快速配准算法研究

A Fast Registration Algorithm of Sar and Panchromatic Images

针对SAR和全色图像配准算法中,数据计算量大、配准时间长的问题,在Powell算法的基础上,提出一种并行优化方法,加快配准速度.引入交叉累积分布熵作为图像匹配测度的基础上,利用生存函数代替密度函数,统计大于某一灰度值的所有直方图信息,保持局部区域灰度信息的连续性,有效改善由噪声和计算误差引起的配准精度的损失.Powell算法可以将求解最优CCRE多参数极值的问题简化为一维极值问题,减少配准过程中迭代的次数,缩短配准时间.为了更快地完成配准,将串行配准算法中大量数据无关的计算部分,分配给多个线程执行,使大块的数据计算处理并行化,从而加快处理速度.实验表明该方法有较好的加速效果.

On the basis of Powell algorithm,aparallel optimization is proposed to reduce the amount of calculation and operation time in SAR and panchromatic image registration algorithms.The paper introduces CCRE（crossing cumulative residual entropy）as measurements of image registration and density functions are replaced by survival functions to calculate the histogram information greater than a specified threshold,and CCRE can effectively improve registration accuracy loss caused by the noise and calculation errors,compared with selecting mutual information.Powell multi-directional algorithm can simplify the multi-parameter extreme-value problem of solving optimal CCRE to one-dimensional extreme-value problem,achieving the purpose of reducing iteration number and shorten registration time.In order to make the registration algorithm faster,aparallel algorithm that assigns amount of independent date computing work to multiple threads is proposed,so that large chunks of data can be processed in parallel,accelerating the processing speed.Experiment shows that this method has well acceleration effect.