Image quality tendency modeling by fusing multiple visual cues

Abstract

Flickr is a photo/video hosting site with over 87 million user. Upload more than 3 million 500 thousand new photos every day at present there are no tools to organize these huge numbers of users aesthetic tendency. Although Flickr allows users manually adding different groups, they are difficult to maintain Updates should be made when new users are added or deleted. In this paper puts forward a series of Flickr users system. Each loop contains similar users aesthetic tendency. We observed: (1) an aesthetic model of thought should be flexible, because of different visual features typically represent different data sets, and (2) Significant differences in the number of photos from different Flickr users stay. In this work, a new probabilistic topic model is proposed describe the aesthetic interest of each Flickr user potential spatial distribution. After that, an affinity graph is similarity is described by aesthetics interests of Flickr users. Obviously, intensive users of Flickr are similar in taste. Thus, these users are divided into different Flickr bounds efficient dense graph discovery. Piping it is proposed that the Flickr bound discovery is fully automatic. Extensive we show that our proposed method is accurate for mine Flickr experiments 60,000 Flickr user community.