图像的SIFT特征二阶检索算法

SIFT features of a second-order retrieval algorithm for images

提出了用图像SIFT特征二阶检索算法实现相似图像检索。首先对图像进行SIFT特征提取,然后利用词袋算法通过K-Means聚类出特征词并建立特征词频表,最后基于词频表通过二阶检索算法实现相似图像匹配。二阶检索分两步:第一步实现特征分布结构的相似检索;第二步根据两张图像对应特征点落于同一特征类的数量与图像自身特征点数的比例来实现图像的精确检索,提高图像检索的准确率。实验结果表明,该方法具有较高的查全率,同时在查询效率上也具有很好的表现。

This paper proposes a second-order retrieval algorithm,which can be used to retrieve similar images.The image SIFT features are first extracted. The frequency table of characteristic words is then built by K-Means clustering and a bag of words algorithm. Finally,based on the word frequency table,similar images can be retrieved by means of a second-order retrieval algorithm. The first step is retrieving the images that have similar distribution characteristics in their structure. The second retrieval step involves accurate retrieval of images according to the proportion of the corresponding feature points that belong to the same class. The experimental results show that this method has both a good recall factor and high query efficiency.