Anotação de pessoas e eventos em fotografias digitais.

Abstract

The popularity of digital cameras has created a new problem: how to store and retrieve efficiently a large number of captured and chaotically stored digital photos in multiple locations without annotation. The photo context assists in search of photographs. The most relevant information for a person remember a photograph are: who are present, where and when was captured. To annotate this information, manual, automatic and semiautomatic mechanisms can be used. The manuals mechanisms did not have many fans due to the costly and tedious process. Content analysis and face recognition are the main strategy to automatic mechanisms, which has only front faces with good results and without occlusions. In the semiautomatic annotation, recommendation of annotations are used to assist the user. In this research, we propose algorithms for the automatic and semiautomatic annotation of people and events in photos. For the annotation of people, we use face recognition for automatic annotation and content and context information for generate suggestions from people for semi-automatic annotation. For the annotation of events, we propose a method for detecting events in personal photo collections and a method for detection of shared events, in which the photographs are captured by different users in the same event. The method to detect shared events aims to improve the search of photographs of an event in a social network, conducting cross-annotation and detection of inconsistencies in the annotations of photographs. The results prove that the weighting and filtering algorithms estimators for semi-automatic annotation of persons in photographs based estimators improve the results of these algorithms. Furthermore, it is possible to detect shared events in a social network using information of who, where and when of the photos.