Image Semantic Understanding In Social Media

Traditional image semantic understanding are trapped in the semantic gap problem. With the rapid development of Internet and popularity of mobile devices, marvelous of social media data floods to the Internet. More and more popular pictures and videos are uploaded or shared to web sites like Flickr and Facebook, image BBS such as Photosig.com, Photo.net and so on. People are drawn to take pictures to share the moment of life. Social media produced a huge amounts of image data and its context information, which will help to a certain extent, or in a particular application across or cut the semantic gap of image understanding. This paper focus on an application of image semantic understanding in social media:placing human in the landscape by analyzing data driven image composition when we are taking photos.Firstly, this paper designs and implements a kind of image composition modeling method based on statistics. This paper models the image composition by fitting a Gaussian mixture model to the saliency map and human position features of images which are handled with principal component analysis. Relation between image feature characters and landscape features is taken fully use in the image composition model, which converting subjective composition rule to objective data model.Secondly, on the basis of the image composition modeling, a photography suggestion system is given based on mobile devices, which aims to help the user place the person to a suitable position and size in the landscape image. Experiments on the paper’s data set show that the proposed image composition model in this paper achieves the desired effects compared to the photography composition rule. The photography suggestion system based on the image composition model in this paper can effectively help users to take photos which follow the photography composition rule to some extent.