摘要
利用Bag-of-feature方法将图像语义近似的转化为一维序列,使用二阶条件随机场将图像的底层语义特征和高层知识关联起来,识别图像内容,并通过对多个初始特征不同的条件随机场推理结果的组合学习来提高图像内容识别的准确率,从而提高分类精度。标准图像数据集的实验证明,该方法有效关联了标签语义,取得了更好的多标签分类效果。
The apphcation of Bag-of-feature method was made to the image language approximate conversion into one dimensional series. And 2na order conditional random fields were used to associate the low level image feature, high level knowledge and to understand the image content. Conditional random fields with different initial feature values were ensemble to acquire higher recognition precision, Simultaneously, classification performance was enhanced. Experiments on benchmark dataset demonstrate that our method effectively associates label semantics and gains better performance on multi-label image classification.
出处
《辽宁工业大学学报（自然科学版）》
2012年第4期223-226,230,共5页
Journal of Liaoning University of Technology(Natural Science Edition)
