一种基于最小均方差的散斑图像配准算法的研究

Research on a Speckle Pattern Registration Algorithm based on Minimum Mean Square Error Method

一种基于最小均方差理论的散斑图像亚像素配准算法,是按线性估计理论推导目标函数的解析解,利用刚体位移模型和散斑图像仿真模型生成5组40帧长斑图像序列,分别在无噪声干扰和加性高斯噪声干扰下对算法进行实验验证。实验结果表明,在无噪声干扰下,图像序列的位移轨迹具有较好的线性度,且系统误差和随机误差较小。在不同噪声水平干扰下,当信噪比为10dB时,该算法失效;而当信噪比不低于20dB时,噪声对配准结果的影响随着信噪比的增加将逐渐减弱。

Based on the theory of the minimum mean square error, one kind of sub-pixel speckle pattern registration algorithm is presented and investigated. According to the orthogonality principle for linear estimators, the analytic expressions of the object function are developed for image registration. 5 columns of the speckle pattern series with the length of 40 frames are generated utilizing the rigid body displacement model and modeling method to simulate speckle patterns. Furthermore, under the conditions of the non-noise and different additive Gaussian noise degradations, a large number of experiments a~ implemented to inspect the effectiveness of the proposed algorithm. The experiment re- suits show that the registration trace of the image series presents the favorable features including linearity, system and random error without the noise disturbing, while the presented algorithm does not work if SNR is less than 10dB and the influence of the registration resolutions from the noise disturbing will be gradually decreased with the increasing of SNR.