基于多特征融合与PLSA-GMM的图像自动标注

摘要

为减少图像检索中图像信息的缺失与语义鸿沟的影响,提出了一种基于多特征融合与PLSA-GMM的图像自动标注方法.首先,提取图像的颜色特征、形状特征和纹理特征,三者融合作为图像的底层特征;然后,基于概率潜在语义分析(PLSA)与高斯混合模型(GMM)建立图像底层特征、视觉语义主题与标注关键词间的联系,并基于该模型实现对图像的自动标注.采用Corel 5k数据库进行验证,实验结果证明了本文方法的有效性.%In order to reduce the impact of semantic gap and lack of image information,a way to improve the image automatic annotation with multi-feature fusion and PLSA-GMM is proposed.Firstly,it is used as the bottom feature of the image with the color feature,shape feature and texture feature of the image.Then,the relation among the low-level feature,the visual semantic topics and the key words is built based on the probabilistic latent semantic analysis(PLSA) and the Gauss mixture model(GMM).Finally,the images can be annotated based on the introduced model.According to test on the widely used database Corel 5k,the results show that the new scheme gives the better performance.