Um ambiente para indexação e recuperação de conteúdo de vídeo baseado em características visuais

Abstract

This work is centered in the area of video information retrieval based on its visual characteristics. It proposes an environment for automatic summarization and indexing of digital videos in order to support the operations of query based on visual content in a video repository. The technique of automatic summary is based on a benchmark avaliation that consider balance between key frame losing and redundance based on visual aspects. As a result, it is possible to obtain a very low representation of the video through key frames, which store enough information about the visual characteristics of the video content. From this small set of key frames, statistics measures and an entity called image signature are taken to characterize the visual content of the videos. The image signature has the ability to represent each key frame in a even more compact way, while maintaining the visual aspects that characterize it, providing a powerful tool for indexation. The approach described was applied to a repository containing more than 34 hours of journalistic videos, thus compressed by about 1.25% of the total number of frames. This reduced representation was then used to index the video contents, providing the basis for various experiments and analysis of performance for the recovery of visual content in videos with the application of various techniques implemented in this work.