Multi-Script Video Caption Localization Based on Visual Rhythms

Localization of video caption plays an important role in information retrieval multimedia applications. In this work, we present and evaluate a novel method for localizing captions using visual rhythms, which enable the representation analysis specific feature throughout time. We build rhythms from text location maps produced by general localization methods that are far more common literature than caption-oriented ones. Then, process properly to keep only captions, generating masks. To meet need standardized large dataset, constructed new one, where with thirteen different scripts added frames, total 221 videos ground truth. Experiments demonstrate our achieves competitive results when compared other approaches.