基于区块位置与加权视觉语义映射的语义图像检索

Semantic Image Retrieval Based on Region Location and Weight Visual Semantics Mapping

提出了一种新的图像语义映射方法WVS-RSSVM,采用自适应的NCut分割方法自动发现并图像中的区域,提取出每个区域包含了位置信息的特征,达到消除一定歧义的目的。并将这些区域采用加权的方式映射为视觉语义空间中的一个点,然后通过SVM分类的方法进行语义学习,实现对图像的语义标注。并且以SVM分类时点到边界的距离作为该点属于某个语义的隶属度,实现检索的排序。实验结果表明,该方法对表达图像的主要语义以及发现有歧义的区块代表的语义,有很好的效果。

A novel image semantics mapping method named WVS-RSSVM is proposed, which employ self-adapting NCut to segment image in- to regions. The region features including location information is extracted to disambiguate of the regions. The image is mapped into the visual semantics space by weighted mapping method. Then its semantics are learned by SVM to annotate the image. The distance between the image feature vector and the SVM＇s hyperplane is used as the degree of the image belongs to the class. It＇s proven that in this method ,the ambiguity can be eliminated to a certain extent.