基于三角网下的仿射不变几何约束的图像匹配算法研究

Study on affine invariant geometric constrained image matching algorithm based on triangle mesh

针对当前图像匹配方法的鲁棒性差、误配率较高及效率较低等不足,提出了基于三角网下的仿射不变几何约束的图像匹配算法。在尺度空间上通过Hessian矩阵对特征点进行检测,利用子块的三角特征与对角特征SURF(speeded up robust features)机制进行改进,用于生成新的特征描述子,并通过定义阈值评估策略对图像特征点进行匹配,从而生成了初始匹配点;然后,引入Delaunay三角网,对初始匹配点进行聚类,以获取匹配三角形,将三角形以外的无效特征点剔除;最后,引入仿射不变几何约束,对匹配三角形进行细化,通过细化的匹配三角形获取最终的匹配特征点,有效剔除误配点,进一步提高配准精度。仿真结果表明,与当前图像匹配算法相比,所提算法具有更好的鲁棒性,且其具有更佳的匹配精度与效率,有效剔除了误配点。

In order to solve these drawbacks such as the poor robustness and large matching error and low efficiency of the current image matching method in image matching, this paper proposed an affine invariant geometric constrained image matc- hing algorithm based on triangle mesh. First, it used the Hessian matrix determinant in the scale space to detect feature points, and used the triangular and diagonal feature of the block to generate feature descriptor by generation method of SURF （ speeded up robust features）. Then it defined the threshold evaluation strategy to generate initial matching points for matching feature points. It introduced the Delaunay triangle mesh to cluster the initial matching points and obtained the matching trian- gle and removed invalid feature points beyond the triangle. Finally, it used the affine invariant geometric constraint to refine the matching triangles, and obtained the final matching feature points by the refined matched triangles, effectively eliminated the wrong matching points, and further improved the registration accuracy. The simulation results show that, compared with the current image matching algorithm, this algorithm has better robustness and better matching accuracy and efficiency, effec- tively eliminates the wrong matching points.