自适应窗口仿射最小二乘匹配方法

为了解决仿射最小二乘匹配方法在未知初始参数条件下无法自适应选择匹配窗口的问题,提出了一种基于局部形状估计的自适应窗口最小二乘匹配方法。利用LOG(Laplacian of Gaussian)算子检测特征点以及特征点所在的尺度;在特征点邻域内根据特征尺度计算局部二阶矩矩阵,用二阶矩矩阵估计特征点局部的形状特征;根据特征点局部的形状自适应选择初始匹配窗口进行仿射最小二乘迭代。实验结果表明,该方法在较大仿射变形下能够准确的选择有效匹配区域,避免因窗口内容差异造成的迭代不收敛问题。

An adaptive window affine least squares matching method

Due to the affine least squares matching method is unable to select the matching window adaptively without the initial parameter, we propose an adaptive window least squares matching method based on local shape estimation. First, we use of LOG (Laplacian of Gaussian)detector to detect feature points and characteristic scale; second, we calculate the second-order moment matrix in the neighborhood of feature points according to the characteristic scale; Finally, the local shape estimation is applied to the adaptive selection of the initial matching window for affine iterative least squares method. The experiment results show that, the method can accurately select effective matching area under the larger affine deformation, avoiding iterative convergence problem caused by the window content differences.