基于曲率的全仿射曲线图像配准

为了解决视点变化造成曲线图像匹配和识别困难的问题,利用全仿射模型建立仿射变换图库,在图库样本中通过曲率信息找出与目标样本相关性最优的样本,从而达到曲线图像匹配和识别的目的。基于曲率的全仿射曲线图像匹配的方法, 首先根据全仿射模型对样本图像建立样本库,对样本库中每一个样本通过等间隔偏移进行采样,然后对采样后的子样本建立曲率样本子集,将目标图像与样本子集中的每一个样本进行快速最近邻搜索算法相似性匹配；最后通过匹配结果,找到最优匹配样本。实验结果表明,基于曲率的全仿射曲线图像配准具有较高的成功率,并且与传统算法相比具有更多的优点。

Full Affine Curve in Curvature Scale Space Image Registration

To solve the problems that viewpoint change of curve image results in difficult matching and recognition,we used full affine transformation model library in the gallery to find the optimal sample correlated with the target by the curvature information,so as to achieve the matching and identification of curve image.In this paper,based on image matching method of curvature full affine curve, a sample library of sample images was established according to the full affine model.In a sample library,each sample is sampled by equal interval offset sampling.Then curvature subset of samples is established for the sub-samples.A fast nearest neighbor search algorithm is carried out using the target ima-ge and the sample subset for similarity matching.Finally the optimal matching sample is found by matching results.Experimental results show that the full affine curve image registration based on curvature has a high success rate,and has more advantages compared with traditional algorithms.