针对全空间仿射不变(ASIFT)算法在解决大视角场景下的图像匹配问题时存在稳健性差、匹配效率低的缺点,提出了一种新的图像匹配算法。采用非线性扩散滤波代替高斯线性滤波对图像进行预处理,以提高检测到的特征点的稳健性;在图像模拟变换过程中采用掩模算子对有效区域进行标记,以提高特征点检测效率。根据视角模拟变换原理,提取特征点不同视角变换下的邻域信息,提出了多视角最近邻匹配与加权匹配规则,建立了多视角描述子,提升了匹配效率。实验结果证明,相比现有的特征匹配算法,所提算法不仅对视角变化具有很好的稳健性,还提高了图像匹配效率与准确度。

An image matching algorithm is proposed to address the poor robustness and low matching efficiency exhibited by the affine scale-invariant feature transform algorithm used for image matching with a large viewing angle. The proposed algorithm employs nonlinear diffusion filtering to preprocess images instead of Gaussian linear filtering, thereby improving the robustness of the detected feature points. Further, a mask operator is employed to denote the effective region in the image simulation transformation process for improving the detection efficiency of feature points. According to the angle simulation transformation principle, the neighborhood information can be extracted from the feature points at different view transformation angles, and the multi-view nearest neighbor matching and weighted matching rules are proposed to establish a multi-view descriptor, thereby improving the matching efficiency. The experimental results show that the proposed algorithm not only exhibits good robustness when the viewing angle changes, but also improves the image matching efficiency and accuracy compared with the existing feature matching algorithms.