Real-time Video Optical Character Recognition (OCR) Demonstration
Videos contain various types of content (e.g., speech, text, faces, vehicles, and humans), which, if recognized, would be valuable as indexes for information search and retrieval. In particular, videotext, in the form of overlay or in-scene text, is a source of concise, valuable, and often unique information about the scene and about some of the objects in that scene. A familiar example from broadcast news video is the name and location often displayed below an image of a person or other entity such as a building - critical information which is often not mentioned in the speech track or closed captions. Other examples of videotext include text on protest signs and banners and road signs in ground reconnaissance videos; name tags, whiteboard text, and projected visual information in videos of meetings and conferences; and text such as serial numbers on the side of vehicles in surveillance footage.
This demonstration showcases a scalable application of the video OCR capability that is ready for transition to operational environments. The streaming video shown in this page displays the result of our real-time videotext detection and recognition processes on live news feeds, as well as archived videos.