Anotação automática e semiautomática de pessoas em fotografias digitais utilizando eventos compartilhados.

Abstract

Faced with the popularization of smartphones and photographic devices, it is a fact that digital photographs have become part of daily life. Millions of photos are added daily on social networks such as Facebook and Instagram. Faced with this, one has the following question: "How to manage huge photo collections?" For the user, this is not a simple task. To assist in resolving this issue, it is necessary to use photo annotations, which may contain location information, date, time, and content tags. For the user, identifying the people present in the scene is considered the most important information, allowing him to remember the photograph. On the other hand, using the concept of the event to which a photograph belongs is a way of optimizing the organization of photographs. In this work, it is proposed to verify if the concept of shared event, which consists of several people with smartphones and photographic devices participating in the same event, brings some increase in the hit rate in annotating people in digital photographs in an automatic and semi-automatic way. For the automatic people annotation, we used algorithms of face detection and recognition; and for the semi-automatic people annotation, estimators based on the context information of the photographs (such as geographic location, for example) and content (such as detected faces, for example) were used. The experiments performed showed good results, thus demonstrating that the use of the concept of shared event increases the hit rate for both people automatic and semiautomatic annotation of in digital photographs.