Arabic Text Detection in News Video Based on Line Segment Detector

Text embedded in video sequences is very important to semantic indexing and content-based retrieval system, especially for large scale news collection. However, its detection and extraction is still an open problem due to the variety of its size and the complexity of the backgrounds. In this paper, we propose an approach for automatic Arabic-text localization based on a novel method for text-line detection. On the first stage, we use a line segment detector to detect candidate text lines. Then, we propose a word segment identification algorithm based on specific features for Arabic text in order to remove non-text lines. The last stage concerns the text line estimation and text detection in video frames. Experiment results, that we drove on a large collection of video images issued from news broadcasts show the excellent performance of our approach for text detection with different character sizes, directions and styles even in case of complex image background.