改进的抗全仿射尺度不变特征变换图像匹配算法

Improved fully affine invariant SIFT-based image matching algorithm

针对现有匹配算法难以解决图像发生仿射变换特别是发生大视角变换时的有效匹配问题,本文对匹配稳定性较好的尺度不变特征变换(SIFT)算法进行了深入研究和改进。借鉴其模拟和归一化相结合的思想对相机光轴的经度角和纬度角进行模拟并采用SIFT算法进行匹配。结果显示,提出的算法不仅保留了SIFT原有的对仿射变换的抵抗能力,而且对视角变换也有很好的鲁棒性,实现了完全的抗仿射变换。实验结果表明,与传统的SIFT算法相比,本文算法对仿射变换尤其是有大视角改变时有更好的适应性。

In image matching process,the affine transformation is difficult to avoid.In exiting algorithms,Scale Invarian Feature Transform（SIFT） has strong resistance to changes of scale,rotation,translation and illumination changes generated by affine transformation.However,when an image has a view angle change,especially large change,the SIFT is not satisfactory.This paper researches the principle of the SIFT and improves its matching function.The latitude and longitude of camera axis are simulated firstly,and then the images are matched by using the improved SIFT algorithm.Experiments show that the algorithm not only retains the original advantages of the SIFT algorithm,but also been robust to changes of the angle.It has achieved a complete anti-affine transformation.In conclusions,the proposed algorithm is more suitable to affine transformation,especially large angle changes,as compared with SIFT algorithm.