Publication

Abstract : This paper presents a semi-automatic video text detection and removal along with a video completion scheme. In the video text detection stage, accurate edge locations are detected using a new type of image representation called as bandlets. Text locations are found by taking Stroke Width Transform (SWT) of the edge map and are grouped using Connected Components (CCs). Motion analyses of the video frames are done in order to preserve the spatial and temporal consistency of the video. After removing the unwanted text regions, an automatic inpainting scheme is employed to fill in the regions with appropriate data. The proposed inpainting scheme takes advantage of both structural and hybrid inpainting techniques. Evaluation of the approach is done using the user prepared video dataset along with ICDAR competition results. The experimental results demonstrate the effectiveness of both video text detection approach and completion technique, thereby the entire video.