【摘要】 在基于内容的图像检索中,建立图像底层视觉特征与高层语义的联系是个难题.对此提出了一种为图像提供语义标签的标注方法.先建立小规模图像库为训练集,库中每个图像标有单一的语义标签,再利用其底层特征,以SVM为子分类器,“否定概率和”法为合成方法构建基于成对耦合方式(PWC)的多类分类器,并对未标注的图像进行分类,结果以N维标注向量表示,实验表明,与一对多方式(OPC)的多类分类器及使用概率和法的PWC相比,“否定概率和”法性能更好.更多还原

【Abstract】 In the approach of content-based image retrieval, a critical point is to provide maximum support in bridging the semantic gap between low-level visual features and high-level concepts. An annotation procedure for providing images with semantic labels was proposed. The annotation procedure started with labeling a small set of training images, each with one semantic label. An ensemble of support vector machines (SVMs) based on the summation of negative probability, which was constructed by pairwise coupling (PWC), was then got with the content-based image features. It was applied to give the unlabeled image an N dimension label-vector, thus providing users with a conceptualized annotation. The ensemble is better than the one per class (OPC) scheme and the PWC based on the summation of probability.更多还原