基于LMISVM的相关反馈图像检索

Relevance Feedback Image Retrieval Based on LMISVM

相关反馈技术是基于内容图像检索研究的热点。本文针对现有SVM相关反馈中假定相关图像的所有特征为相关这一不完全准确假设,提出了MISVM短期机器学习相关反馈方法。该方法采用多示例学习方法确定图像中每个特征的相关程度来提高SVM的分类准确性;在此基础上,为进一步提高系统反馈速度与准确率,通过保存以前训练好的分类器和反馈样本,提出了基于LMISVM长期机器学习的相关反馈方法。文中提出的两种方法与其它方法进行了比较实验,结果表明该方法优于其它方法。

The relevance feedback technology based on machine learning becomes a focus in image retrieval study. So this paper proposes,a new relevance feedback method based on machine learning, which combines multi-instance learning with SVM. On this basis, to improve the image retrieval speed and accuracy, another new feedback method based on long-term machine learning is presented. Experiments show that the proposed system is not only efficient but also effectiy.