基于SVM与Adaboost的图像检索应用

The Application of SVM and Adaboost in Image Retrieval

语义图像检索已成为解决简单视觉特征和用户检索高级语义之间存在的"语义鸿沟"问题的关键,本文试图提出一种基于SVM和Adaboost集成学习相结合的相关反馈算法。在相关反馈过程中选择最具信息的样本训练支持向量机,可以有效减少相关反馈的次数和所需学习样本的数量,通过两者的互补来有效地提高图像检索的精度。最后提出Adaboost算法对SVM分类器进行加权投票,这样进一步提高了图像检索的性能。实验表明,该方法能较好地解决了图像检索中的小样本选择问题,并能显著提高图像检索的效率和性能。

Semantic image retrieval has been a crux to bridge ＂semantic gap＂ between the simple visual features and the advance se- mantics delivered by an image. In this paper, an image retrieval method combining SVM and Adaboost algorithm is proposed. The proposed approach selects the most informative samples in database to train SVM, it can reduce the feedback rounds and the number of samples effectively, and it can use both advantages to improve the accuracy of image retrieval. At last Adaboost method is pro- posed to integrate studying with SVM, and it improves the image retrieval performance. The experiments show that our method works well in solving the small sample size problem and it can improve the retrieval efficiency and performance consistently under the con- dition of limited training samples.