结合动作视觉概念的复杂查询图像重排序

Image reranking for complex query combined with verb visual concept

从复杂查询中挖掘动作视觉概念,提出面向复杂查询时将动作视觉概念亦纳入考虑的图像检索结果重排序方法。首先从复杂查询中提取动词和名词短语作为视觉概念,然后分别从语义层、视觉层以及两者的交叉形态估计复杂查询与图像之间的相关性,最后根据相关分数完成检索结果的重排序。通过在Google图像搜索引擎上的实验结果证明,针对复杂查询的检索结果重排序,加入动作视觉概念能够更加具体地描述图像的视觉内容。

This paper proposed a reranking approach for complex queries which took the verb visual concepts into account.Giving a complex query,it first detected the noun-phrases and verb-phrases as visual concepts. Then it estimated the relevance scores from three layers,i. e.,the sematic-level,visual-level as well as cross-modality level. Based on the relevance scores,it could obtain a new ranking list. The experimental results in Google image search engine show that this approach can obtain good performance for complex queries on the description of image visual content.