基于文本和内容的图像检索算法

Abstract

为了提高图像检索的效率,提出一种基于文本和内容的图像检索算法。该算法采用稠密的尺度不变特征转换(dSIfT)构造视觉单词的方式来描述图像内容,依据基于概率潜在语义分析(PlSA)模型的图像自动标注方法获取的视觉语义对查询图像进行初步检索,在此结果集上对筛选出的语义相关图像按内容相似度排序输出。在数据集COrEl1000上的实验结果表明,该算法能够实现有效的图像检索,检索效率优于单一的基于内容的图像检索算法。

In order to improve the efficiency of image retrieval, an image retrieval algorithm based on text and content was proposed.This method used Dense Scale-Invariant Feature Transform( DSIFT) feature to construct visual words to describe image content, roughly searched query image according to the visual semantics acquired by automatically annotating based on Probabilistic Latent Semantic Analysis( PLSA) model, then sorted the filtered semantically related images according to the similarity of content.Experimental results in Corel1000 database demonstrate that the proposed algorithm can retrieve images effectively and achieve better performance than the algorithm only based on image content.