摘要： 概率隐含语义分析模型不适用于大规模图像数据集，为此，提出一种基于隐含狄利克雷分配模型(LDA)的图像分类算法。以BOF特征作为图像内容的初始描述，利用Gibbs抽样算法近似估算LDA模型参数，得到图像的隐含主题分布特征，并采用k近邻算法对图像进行分类。实验结果表明，与基于概率隐含语义分析模型的分类算法相比，该算法的分类性能较优。

关键词: BOF模型, 中层语义特征, 隐含狄利克雷分配模型, 隐含主题分布特征, k近邻算法, 图像分类

Abstract: To solve the problem that probabilistic Latent Semantic Analysis(pLSA) model is not suitable for large-scale image dataset, a new image classification algorithm based on Latent Dirichlet Allocation(LDA) model is proposed. It uses Bag-of-Features(BOF) model as images initial description, applies Gibbs sampling to estimate the parameters of LDA model, and gets images distribution in the latent topic space. Images are finally classified by k Nearest Neighbor(kNN) algorithm. Experimental results indicate that, compared with algorithm based on pLSA model, the image classification algorithm based on LDA has more powerful classification performances.

Key words: Bag-of-Features(BOF) model, middle-level semantic feature, Latent Dirichlet Allocation(LDA) model, latent topic distribution feature, k Nearest Neighbor(kNN) algorithm, image classification

中图分类号: 

• TP341.94