Arabic-text extraction from video images

الملخص EN

Abstract Automatic extraction of meaningful objects in video is extremely useful in many practical applications.

Text objects embedded in video contain much semantic information related to the multimedia content.

In this paper, we proposed an algorithm to detect/localize and segment Arabic static artificial texts embedded in video.

Firstly, the edge map of a predetermined region of interest is detected in a single video frame.

Projection profiles are adopted to localize the candidate text regions followed by some filtering rules which are used to filter out non-text regions.

Secondly, since artificial text lasts for a certain time on screen for reasons related to human vision, therefore the temporal information are exploited in order to reinforce the recall and precision rates.

Finally, a thresholding-based method is used for separating text pixels from background pixels and produce a binary text image.

To improve high recognition rate, robust enhancement methods