Video Content Annotation Method Based on Multi-weight Multi-source Transfer Learning

In solving the problem of video content annotation, this paper proposes a Multi-Weight Multi-Source Transfer Learning (MW-MSTL). According to the degree of correlation between the source domain image group and the target domain video, the method assigns different weights to different source domain image groups. Finally, based on the smoothness assumption, a target classifier capable of classifying user videos is trained. and then complete the task of video content annotation. Experiments show that the average accuracy of this method on kodak and Columbia Consumer Video(CCV) database is 42.68% in the absence of labeled data on the target domain. Compared with other methods, it has achieved better labeling effect.