基于Corr-LDA模型的图像标注方法

Image annotation method based on Corr-LDA model

针对现有图像标注方法大多将不同类别的图像置于同一主题空间下进行标注的不足,提出了一种新的图像标注方法,该方法以Corr-LDA模型为基础,将各类图像置于不同主题空间下,并为每个类别学习出适合该类的图像标注模型。在Labelme及UIUC-Sport数据集上的实验结果表明,本文方法的标注性能要优于其他方法。

Most of existing image annotation methods are based on the same topic space.In fact,categories are important information to determine the image annotation words,different classifications of images present different objects.In this paper,a new image annotation method based on Corr-LDA model is proposed.In this method all sorts of images are placed in different topic spaces,and the image annotation model suitable to each category is learned.Experiment results on Labelme and UIUC-Sport datasets show that the proposed method has better performance than other methods.