Annotating videos from the web images

In this paper, we propose a generic framework for annotating videos based on web images. To greatly reduce expensive human annotation on tremendous quantity of videos, it is necessary to transfer the knowledge learned from web images with a rich source of information to videos. A discriminative structural model is proposed to transfer knowledge from web images (auxiliary domain) to the video (target domain) by jointly modeling the interaction between video labels and we-b image attributes. The advantage of our framework is that it allows us to infer video labels using the information from different domains, i.e. the video itself and image attributes. Experimental results on UCF Sports Action Dataset demonstrates that it is effective to use knowledge gained from web images for video annotation.