基于尺度不变Harris特征的准稠密匹配算法

Quasi-dense matching based on scale invariant Harris feature

准稠密匹配是多视图三维重建的重要技术,其性能对重建结果至关重要。针对常用的SIFT算法提取的种子点进行准稠密匹配正确率较低、重建效果不佳的问题,提出了一种基于尺度不变Harris角点特征的准稠密匹配算法。该算法在图像多尺度空间构造尺度不变Harris特征,并采用余弦距离测度对不同视图进行双向匹配。根据稀疏匹配获取种子点,采用最优最先匹配扩散策略进行准稠密扩散,采用局部非极大值抑制策略对匹配结果进行重采样。实验表明,算法提取的种子点既能够体现场景结构信息,又具有尺度不变特性,用于准稠密匹配,能够提高匹配的效果和精度,是一种有效的用于三维重建的准稠密匹配算法。

Quasi-dense matching is widely used in multi-view 3D reconstruction,and it is important for reconstruction results.Aiming at the quasi-dense matches diffused by the seed points extracted from SIFT algorithm were less accurate,this paper proposed a quasi-dense matching algorithm based on scale invariant Harris corners.Firstly,this algorithm structured the scale invariant Harris features in multi-scale space,and bidirectionally matched the feature sets between different views by cosine distance similarity measure.Then it applied the seeds selected from the initial matches in quasi-dense matching algorithms by best and first propagation strategy.Finally,it applied a local non-maximum suppression strategy to resampling the quasi-dense matching results.Experiments show that the seeds extracted by the proposed algorithm can not only reflect the scene structure information,but also have scale invariant characteristics.And for quasi-dense diffusion,the matching effect and accuracy can be improved,and it is an effective quasi-dense matching algorithm for 3D reconstruction.