利用位置信息加权词汇树的图像检索

Image Retrieval Using Weighted Vocabulary Tree with Location Information

针对基于SIFT特征匹配的图像检索方法忽略空间位置信息的缺陷,提出了基于空间位置信息加权词汇树的图像检索方法。利用分层词汇树将图像SIFT特征量化为视觉单词,将图像匹配转换成视觉单词权值向量匹配。由于单纯的视觉单词权值向量匹配忽略了视觉单词之间的相互位置影响,生成SIFT点的空间位置影响信息,根据SIFT点与视觉单词之间的从属关系,将SIFT点的空间位置影响信息聚类为视觉单词之间的位置影响,将视觉单词的空间位置信息作为加权系数,加入到视觉单词的权值匹配中去,细化特征之间的匹配得分,通过相似度排序检索相似图像。实验结果表明,该算法能有效提高图像检索的精确度。

Aiming at the positional information ignorance problem in image retrieval based on SIFT feature matching, an image retrieval approach using weighted vocabulary tree based on the spatial location information was proposed. The vocabulary tree is used in the method to quantify SIFT features as the visual words, converting the image match into the visual words＇ weight vector match. Because only visual words＇weight match ignored the mutual position effect, SIFT points＇spatial location information was generated, and was classified into the position effect of visual words according to affiliation between SIFT points and visual words. The spatial location information of the visual words was used as the weighting factor of weight vector matches, refining the matching score between features. Similar images were retrieved by similarity sort. The experimental results show that the algorithm can effectively improve the accuracy of the image retrieval.