一种新的宽基线图像匹配方法

Wide baseline image matching based on a new local descriptor

在宽基线图像匹配中,图像存在3维视角、尺度、旋转和灰度差异.为此,构造了一种新的基于局部二值模式直方图傅里叶特征的特征描述符,并通过对传统宽基线图像匹配算法框架中不同部分算法的对比分析,提出了一种新的宽基线图像匹配方法.首先,提取基准图像和实时图像中具有尺度和仿射不变性的最稳定极值区域,并利用新的特征描述符对这些区域进行图像旋转和灰度不变性描述;然后,根据近邻欧氏距离比值准则提取两图像中匹配的最稳定极值特征区域对;最后,利用顺序抽样一致性算法剔除误匹配特征区域对,估计两图像的外极几何关系,得到匹配结果.仿真结果表明,新算法能够适应待匹配图像间较大的3维视角、尺度、旋转和灰度差异,实现稳定的宽基线图像匹配.

In order to overcome the differences in 3D viewpoint,scale,rotation and grayscale,and achieve stable wide baseline image matching,a new feature descriptor based on Local Binary Pattern Histogram Fourier features is constructed and a new wide baseline image matching approach is proposed according to comparison and analysis.First,Maximally Stable Extremal Regions（MSER） of the reference image and real-time image which are scale and affine invariant are extracted,respectively.Second,rotation and grayscale invariant descriptors are constructed.Then the matching MSER features of the two images are extracted based on the nearest neighbor Euclid distance ratio strategy.The epipolar geometry of the two images is estimated according to the Progressive Sample Consensus（PROSAC） algorithm.Simulation results show that the proposed method is robust to changes in 3D viewpoints,scale,rotation and grayscale,and can achieve stable wide baseline image matching.