A Dataset for Arabic Text Detection, Tracking and Recognition in News Videos - AcTiV (AcTiV)

Ground Truth

Global xml file

2016-03-16 (v. 1)

Contact author

Oussama Zayene

SAGE Lab, University of Sousse, Tunisia & DIVA Group, University of Fribourg, Switzerland



This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported License.



The annotation process consists of two different levels:

The global annotation, which concerns the entire video, is performed manually thanks to a user interface. We first open a video clip, and then we draw a rectangle for each static text. Once a textbox has been selected, a new set of information is created. It contains the following elements:

  • Time stamps for its apparition interval: start/end frame.
  • Rectangle’s attributes: (x, y) coordinate, width, height.
  • Content data: text, text color, background color, background type<transparent, opaque>.

Dynamic text is formed by continuous scrolling series of tickers. To annotate this kind of text, we noted for each ticker: its content, the first frame where the ticker appears and the initial offset in the first frame which is estimated using a virtual line.

The local annotation at the frame level is done automatically according to the information contained in the global metafile. For more details please reffer to our work (Zayene et al. IPTA 2014)


No comments on this dataset yet.

Add your comment

In order to comment on a dataset you need to be logged on
Register Now!


In order to rate this dataset you need to be logged on
Register Now!