一种SIFT特征权重排序的多图关联图像检索方法

A Multi Image Retrieval Method for SIFT Feature Weights Sorting

多图检索算法是查询图与其内容相关图像联合进行的图像检索算法。在之前的多图检索算法研究中,大量噪声特征被引入到最终的空间一致性度量,从而导致检索精度无法提升,所以这里提出一种基于SIFT特征权重排序的多图关联图像检索方法。将查询图像和本地存储的一些内容相关的图像进行联合,并从中筛选出最具有代表性的特征点,而后对这些特征点进行权重的计算和排序,以达到更高的检索精度。实验证明这种方法大大减少了特征点的数量,增加了图像检索的健壮性。

The multi-image retrieval method is an image retrieval method based on the combination of query image and the images which are related to it.However,in study of multi-image retrieval method,a large number of noise features were used in spatial consistency measurement,which led to low accuracy of retrieval.This paper proposes a multi image retrieval method based on SIFT feature weights sorting.Associate the query image with the images which are related to it,and then select the most representative feature points.Finally,calculate and sort the weights of these feature points.The method greatly dwindles the number of feature points and enhances the robustness of image retrieval.