基于显著区域和pLSA的图像检索方法

摘要

Because image retrieval method using global features is largely affected by background,presented a novel image retrieval method based on salient region and pLSA. In this approach, salient regions are first detected through spectral residual and multi-resolution a-nalysis, then color and texture features of those salient regions are quantized into a dictionary of visual words by K-mean clustering, after that each image is represented by a bag of visual words. Finally, by exploiting probabilistic latent semantic analysis, achieve the latent semantic feature, which can be used to construct a SVM model to fulfill the image retrieval. Compare the method proposed with the global-based image retrieval method, the experimental results show that the accuracy of image retrieval method based on salient region is higher than the other one.%由于利用全局特征的图像检索方法在很大程度上受到背景的影响,提出了一种基于显著区域和pLSA相结合的图像检索方法.该方法首先通过谱残差和多分辨率分析提取图像的显著目标区域,其次计算所有图像显著区域的颜色和纹理特征并利用K-均值聚类生成视觉词汇表,然后将每幅图像表示成若干视觉词汇的集合.最后利用概率潜在语义分析(pLSA)来提取区域潜在语义特征,并使用该特征构建SVM分类器模型进行图像检索.将本方法和基于全局特征的图像检索方法比较,实验结果表明,基于显著区域的图像检索结果更加准确.