Automatic focus personage identification in multi-lingual news image

This paper presents a novel method for automatic focus personage identification in multi-lingual news image. Other existing methods resolved the problem of person recognition in single-lingual news images, but there is no proposed method to handle the problem of multi-lingual character identification. This method incorporates the complementary and commonality information of multi-lingual news to identify person. It includes two main steps. Firstly, positive training samples are obtained by clustering ensemble approach. Secondly, recurrent convolutional neural network (RCNN) is applied to train model and classify images. The experiment is performed on the news data set, which consists of half twenty thousand news picture-caption pairs from Google image search and Baidu image search. The proposed method yields a much better performance than other methods.