为了实现准确、快速的图像匹配, 从角点检测与描述子两方面入手, 提出了一种基于CSS角点检测的匹配算法。首先, 在曲率尺度空间下, 检测图像在不同尺度下的角点并剔除不稳定角点; 其次, 基于曲率对图像轮廓描述的精确性, 以特征点为中心划分3×4的子邻域, 计算子邻域内轮廓曲线点的高斯加权曲率等4维向量特征, 建立48维描述子, 由于CSS角点检测包含曲率计算, 因此生成描述子时避免了曲率的二次计算, 提高了匹配速度; 最后,提出一种“二进制距离”方法对描述子进行匹配, 进一步优化匹配速度。通过实验证明, 在保证精度的情况下, CSS快速匹配算法大幅度缩短了匹配时间, 对旋转、亮度变化具有较好的匹配效果。

In order to realize accurate and rapid image matching, this paper proposes a corner matching algorithm based on corner point detection and the descriptor. Firstly, in the curvature scale space, the corner points of the image at different scales are detected, and the unstable corner points are removed.Secondly, based on the accuracy of the curvature describing the contour of the image, the neighborhood is divided into 3×4 sub-neighborhoods taking the character points as the center, the 4-dimensional vector features in the sub-neighborhood are calculated, such as the Gaussian weighted curvature of the contour curve points, and a 48-dimensional descriptor is built.Since the CSS corner point detection includes curvature calculation, the quadratic calculation of the curvature is avoided when generating the descriptor, and thus the matching speed is improved. Finally, a binary distance method is proposed to match the descriptors to further optimize the matching speed. Experiments have proved that, with the assurance of accuracy, the CSS fast matching algorithm can shorten the matching time to a great extent and achieve satisfying matching results under rotation and illumination changes.