LMA和对数极坐标变换的匹配算法在景象匹配制导中的应用

Matching algorithm using LMA and log-polar transform in scene matching guidance

针对成像制导过程中对实时性和准确性的要求,通过对存在旋转和尺度变化的航拍照片的匹配进行研究,提出了一种时数极坐标变换 LMA(Levenberg-Marquardt Algorithm)匹配算法.在 LMA 的匹配迭代计算过程中使用8个参数,匹配时先对基准图和实时图按多分辨率金字塔法进行对数极坐标变换,利用对数极坐标变换的旋转和尺度不变性进行匹配,提高子像素的匹配精度,使需匹配照片的像素成最佳排列.迭代过程中通过消除 Hessian 矩阵,提高匹配的实时性.这种对数极坐标变换图像匹配很有实用价值,对航拍照片进行匹配测试,证明了这种方法的有效性.

According to research of the log-polar transform of airscape matching in the spatial domain, Levenberg-Marquardt Algorithm(LMA)with superior performance is demonstrated in this paper.Eight parameters of the perspective transformation are used in the registration process.Firstly,multi-resolution pyramid for both reference and target images is used in matching.Secondly,the rotational and scaling property of log-polar image transformations is used.In this way,the subpixel accuracy is improved by use of LMA,and the pixels of two input images are aligned in the best way.Extensive testing is performed on uncalibrated real images and an array image pairs.As the experimental results shown,this approach is efficient and can deal with transform.Therefore,the problem of transform estimation in log-polar images is tackled in this work.