Text area detection from video frames

Abstract

Text area detection from video frame is an essential step for Video OCR. The key problem is the complex background of the video frames. This paper proposes a novel approach to this problem. First, we use the vertical edge information to detect candidate text areas. The horizontal edge information is then used to eliminate some of the false candidates. Finally, shape suppression technique is applied to further refine the results. Experimental results have shown the proposed approach is very effective in text area detection.