Automatic Video Indexing and Retrieval System for Turkish Videos

Abstract

A continual increase in multimedia data needs a sophisticated automatic video indexing and retrieval system based on its content. This paper exploited statistical feature along with some morphological operations to detect horizontally aligned artificial textual fields in Turkish video frames, which are then extracted for content-based video indexing and retrieval system. First, projection analysis was performed to find the edges in the images and then morphological operations to convert the textual regions in images into lines. Later, false positives were eradicated by geometrical constraints and heuristics-based method. The detected candidate text regions were fed to optical character recognition (OCR) system to recognize and output the text. Finally, the recognized words were stored in database as keys for automatic content-based video indexing, which can be retrieved through provided web interface. For evaluation, a ground-truth preparation software is prepared to manually localize the text in images. Experimental results showed that our proposed method performed well on Turkish videos with overall f-measure of 95%.