Yesterday we showed how Google's Cloud Video API was able to seamlessly identify onscreen Arabic text in an ABC Evening News broadcast and transcribe it alongside the English text in the same frame. Another example from the same broadcast demonstrates the API's ability to recover even severely degraded text. The image at the top of this post is a single frame from the show's coverage of the Deepwater Horizon disaster, showing footage from a remote controlled submersible.

At the top of the image is a computer generated textual overlay with information about the image. While legible, this text is blurry, has severe color distortion with yellows and greens and other colors across the text, uses a font with severe aliasing and is further reduced by video stream compression. Yet here the OCR was able to recover much of the onscreen text. For a frame near the one above it reported "Subsea 7 E 1202246 92 N 1843163155 14/87/10 10431633 12022 Alt 4892.9 Herc 06 Plune Monitoring Hdg LIVE ET GULF OF MEXICO abc". While not perfect, it matches pretty well what a human observer would likely recover from the footage and actually in places exceeds human legibility from the original footage without manual digital enhancement.