Joint Subtitle Extraction and Frame Inpainting for Videos with Burned-In Subtitles

Subtitles are crucial for video content understanding. However, a large amount of videos have only burned-in, hardcoded subtitles that prevent re-editing, translation, etc. In this paper, we construct deep-learning-based system the inverse conversion burned-in subtitle to file and an inpainted video, by coupling three deep neural networks (CTPN, CRNN, EdgeConnect). We evaluated performance proposed method found learning achieved high-precision separation frames significantly improved inpainting results compared existing methods. This research fills gap in application reconstruction is expected be widely applied re-editing with subtitles, advertisements, logos, other occlusions.