基于视觉和语义的图像检索技术研究

Research on Image Retrieval Technology Based on Vision and Semantics

对图像检索技术进行了研究,提出了一种基于视觉和语义的图像检索算法。首先使用稠密的尺度不变特征转换构造视觉单词的方式来描述图像的视觉内容,然后依据一种基于概率隐语义分析的自适应不对称学习方法去融合并学习视觉模态和文本模态信息得到的语义特征对查询图像进行初步检索,最后在此结果集上对选出的语义相关图像按视觉内容相似度排序输出。通过实验表明,利用视觉和语义的算法能够提高图像的检索效果,具有更好的检索性能。

The technology of image retrieval is studied, and an image retrieval algorithm Based on vision and semantics is pro- posed.The first use of dense scale invariant feature transform structure of visual Words to describe visual content of image, then the adaptive asymmetrical learning method Based on probabilistic analysis to integrate and learn the semantic mode and text mode information, the query image is initially retrieved, finally, the selected semantic related images are sorted out according to the simi- larity of the visual content in the result set.Experiments show that the algorithm can improve the retrieval effect and improve the retrieval performance.