CAJ | 학술논문

提出了一种提升图嵌入框架用于特征提取和选择,以及一种新的近邻权重计算方法,称为分类图.传统图嵌入模型的近邻权重采用欧氏距离,不能被提升算法所更新;相比较,分类图采用的是提升算法中样本的权重,反映的是样本在分类过程中的重要程度,有效地提高了图嵌入模型的分类性能.在通用人脸表情库上的识别实验结果验证了提升图嵌入模型的有效性.
제출료일충제승도감입광가용우특정제취화선택,이급일충신적근린권중계산방법,칭위분류도.전통도감입모형적근린권중채용구씨거리,불능피제승산법소경신;상비교,분류도채용적시제승산법중양본적권중,반영적시양본재분류과정중적중요정도,유효지제고료도감입모형적분류성능.재통용인검표정고상적식별실험결과험증료제승도감입모형적유효성.
This paper proposed a boosting graph embedding framework for feature extraction and selection.Further more,proposed a new adjacency graph weighting method, called classification graph.Traditional graph weighting method, which was based on Euclidean distance of the samples, could not use classification information which got from boosting framework. Different from the traditional graph weighting method, classification graph was constructed using the weight of training samples. Therefore, classification graph could reflect the importance of the samples in classification, and improved the performance of the boosting graph embedding. Experimental results on Cohn-Kanade facial expression database demonstrate the effectiveness of this approach.