Linking names and faces by person-based subset clustering

In this paper we address the challenge problem of linking names and faces from online news. Developing accurate technologies for linking names and faces is valuable when retrieving or mining information from images collections. Here, we propose a novel method called Person-based Subset Clustering. It divides into four steps. Firstly, we detect names in the captions. Secondly, for the target name, we retrieval all images with the same name. Thirdly, each image is normalized by calibrating two facial feature points. And then we extract the Local Gabor Binary Pattern Histogram Sequence as the local visual features to represent each face image, and pair-wise distances are calculated. Finally, several clustering methods are applied to clustering the faces. Experiments show the effectiveness of our method, which is based on a data set consisting of approximately 690 news pictures and captions from Yahoo News.