基于爬山搜索的高斯模糊不变SIFT算子

摘要

SIFT operator is difficult in feature matching in Gaussian blur environment.Aiming at this problem,we proposed a Gaussian blur invariant SIFT operator (GI-SIFT)which is based on resampling in deformation space of object image.First,we built the Gaussian blur model of the clear object and re-sampled the model parameters to reconstruct complete deformation space of the object image.Secondly,we introduced subsampling and hill climbing approaches to construct the subsampling deformation space of the object image,and rapidly searched the current peak value in subsampling space with large sampling step,and made curve fitting in peak neighbourhood to quickly find the optimal matching.Experimental results showed that the proposed algorithm can well match Gaussian blur object,at the same time it also greatly improves the efficiency of objects feature matching.%针对SIFT算子对于高斯模糊环境下的特征匹配困难，提出基于目标图像形变空间重采样的高斯模糊不变SIFT算子GI-SIFT（Gaussian Invariant SIFT）。首先构建清晰目标的高斯模糊模型，重采样模型参数重建目标图像完备形变空间；其次，引入降采样与爬山法，构建目标图像的降采样形变空间，在降采样空间中以大采样步长快速搜索当前峰值，对峰值邻域进行曲线拟合，快速找到最优匹配。实验结果表明，所提算法不仅对高斯模糊目标能较好匹配，同时较大提升了目标的特征匹配效率。