参数优化模型的正交逼近算法

摘要

Image registration is transformed an unconstrained optimization problem in image mosaic. Parameter optimization model algorithm is proposed by means of quadratic form theory-orthogonal approximation algorithm. According to properties of best-function, a direct method is adopted without calculating derivative or gradient, only calculating function values. The advantages are eliminating of taking up large computer memory, only using a matrix to store information. Simulation experiments show the algorithm has advantages of fast convergence, result precisely and good practicality.%在图像拼接中,该文将图像配准转化成一个无约束最优化问题,运用2次型理论提出一种新的参数优化模型算法正交方向逼近法.该算法根据参数矩阵最优函数特有的性质,无须计算函数的导数或梯度,仅须计算函数值即可.其优点是克服了以往算法所要求占用大量计算机内存的缺点,仅用一个矩阵存储数字信息.通过仿真实验证明:该算法收敛速度快,结果精度高,具有较高的实用价值