Nowadays, the large volume of multimedia data available motivates the study of techniques for manipulation and indexing of images and videos. In this context, the goal of this research project is to investigate the use of visual dictionaries in order to characterize the visual properties of images in applications related to Information Retrieval and Multimedia Classification. The method based on visual dictionaries appears as one of the most popular techniques to describe images. However, this approach does not consider any spatial information to describe images and objects. The purpose of this project is to obtain a description semantically richer of the image visual content using graphs to incorporate the spatial distribution of visual words to the visual dictionary model. This use of graphs is motivated by the invariant behavior of these structures under various transformations of objects and images, such as rotation, translation.The contributions of this project fit into the context of one of the Great Challenges in Computing of SBC: Information Management in Large Volumes of Distributed Multimedia Data.
News published in Agência FAPESP Newsletter about the scholarship: