We study the problem of automatically discovering the interpersonal closeness between people from a personal photo album in this paper. The work is divided into two parts, the first part is the automatical detection of person figures and the identity annotation, while the second part is the analysis of the interpersonal closeness between identities. The detection of person figures is solved by the cooperation of a face detector and a person detector. Similarities between pairs of person figures are calculated by face verification and color descriptor matching, and a greedy stratagem is used for identity assignment. The second part of the work is a ranking process in which for each identity the other identities are ranked by his(her) closeness to them. The algorithm aims at analyzing and ranking interpersonal closeness according to the image information extracted from the album. The image information used by our method includes co-appearing frequencies of persons, order distances between person figures and the number of persons in each photo. We build a new database for this work. Experiments on our database show that our method is effective especially for finding the closest persons. The closeness rankings can be used to do further interpersonal relationship analyses such as discovering subgroups and finding the most active members.
展开▼
