CAJ | 학술논문

针对半监督特征选择算法进行了研究，采用有标签图像和无标签图像的半监督特征选择方法来提升网络图像标注的性能。基于二阶Hessian 能提出一个新的半监督特征选择方法，该方法具有更好的局部拓扑结构保持特性和推断能力，从而能够克服基于图拉普拉斯半监督学习方法的缺点。将所提出的半监督特征选择算法应用到网络图像标注任务中，在两个大规模网络图像数据库上进行了实验，结果表明Hessian半监督特征选择方法优于拉普拉斯半监督特征选择方法，适合大规模网络图像标注。
침대반감독특정선택산법진행료연구，채용유표첨도상화무표첨도상적반감독특정선택방법래제승망락도상표주적성능。기우이계Hessian 능제출일개신적반감독특정선택방법，해방법구유경호적국부탁복결구보지특성화추단능력，종이능구극복기우도랍보랍사반감독학습방법적결점。장소제출적반감독특정선택산법응용도망락도상표주임무중，재량개대규모망락도상수거고상진행료실험，결과표명Hessian반감독특정선택방법우우랍보랍사반감독특정선택방법，괄합대규모망락도상표주。
This paper researched and applied semi-supervised feature selection method,which exploited the labeled data and unlabeled data simultaneously,to improve the performance of Web image annotation.It proposed a novel semi-supervised fea-ture selection method based on the second-order Hessian energy,which could preserve local topology better and had good ex-trapolation capability to overcome the drawbacks of methods based on Laplacian.It applied the proposed method to Web image annotation task and performed extensive experiments on two large-scale Web image datasets.The results show that the proposed method is superior to methods based on Lapacian and suitable for large-scale Web image annotation.