中文摘要：

两幅图像中相应特征点邻域窗口之间的单应映射可以用仿射变换模型来近似。本文首先通过奇异值分解给出仿射变换矩阵4个自由度的几何含义，然后将其分解为一个相似变换矩阵和一个旋转的准单位矩阵（Rotated Quasi—Identity Matrix）的乘积，即在基于相似变换模型匹配的基础上再用基于仿射变换模型的迭代算法对相应特征点精确定位。针对相似变换中初始旋转角度的难确定性，在初始匹配中提出基于亮度最速下降方向的对齐方法，而在引导匹配阶段提出基于相应极线方向的对齐方法，这两个策略不仅提高了算法效率，还能为进一步的仿射迭代提供良好的初值。在得到最优仿射变换参数之后，实现了对相应特征点定位误差的精确补偿及其邻域窗口的透视矫正。最后通过真实图像的实验以及和现有算法的比较验证了本文算法的可行性和精确性，并给出了相应的实验数据和结果。

英文摘要：

The homography between corresponding matching windows can be approximated by an affine model. Traditional optical flow based iterative methods can track feature points in video sequence accurately, but good initial parameters must be provided, and it is impossible to apply such methods to wide baseline images mechanically. In this paper, meaning of the four parameters of affine model was given first,and then the affine matrix is decomposed into a similar matrix and a rotated quasi-identity matrix,which uses a similar model followed by an affine model using iterative optimal method. Since the search of initial rotation of corresponding matching windows is nonlinear,this paper proposes a fast method based on alignment of the fastest descend directions of intensity （FDDI） within corresponding .matching windows in the un-guide stage,and a fast method based on alignment of corresponding epipolar lines （CEL） ,which not only give good initial parameters,but speed up the algorithm. After getting the optimal affine parameters,the location error of corresponding feature points were compensated, and the deformations of corresponding feature windows were rectified. Results of real test images were given ,which strongly demonstrate the feasibility and accuracy our method.