摘要： 为了避免图像分割，并提高图像标注精度，提出一种基于典型相关分析(CCA)和高斯混合模型(GMM)的自动图像标注方法。利用CCA对图像的全局颜色特征与全局局部二值模式(LBP)纹理特征进行特征融合。使用融合后的语义特征，对每一个关键词建立GMM模型来估计单词类密度，从而在特征子空间中得到每个单词的概率分布。采用贝叶斯分类器确定每个标注词和测试图像的联合概率，运用词间语义关系优化标注结果。实验结果表明，使用该方法后的图像标注性能有了较大程度的改善。

关键词: 典型相关分析, 特征融合, 高斯混合模型, 类密度, 图像标注, 子空间

Abstract: In order to improve the accuracy of image annotation without using image segmentation, this paper proposes a new automatic image annotation method based on Canonical Correlation Analysis(CCA) subspace and Gaussian Mixture Model(GMM). Do feature fusion between global color feature and global LBP texture feature by using CCA, establish GMM model to estimate the class density probability of every word using the fused feature, and obtain the probability distribution in feature subspace for each word. The joint probability of each tag with respect to the test image is calculated by Bayesian classifier. And the annotation results are refined using semantic correlation between annotation words. Experimental results show that the annotation performance is greatly improved using the proposed method.

Key words: Canonical Correlation Analysis(CCA), feature fusion, Gaussian Mixture Model(GMM), class density probability, image annotation, subspace

中图分类号: 

• TP391.41