Recognition of Characters from Streaming Videos

Over the past few years, Video has become one of the prime source for recreation, be it Television or Internet. Television brings a whole lot of professionally produced video content (International or local, sports or educational, news or entertainment) to the home for the masses. Similarly, Internet hosts a whole lot of video content uploaded by other users. Understanding the context of the video automatically can open up avenue for a lot of valueadded applications. Now, what do we mean by understanding the context? Video context is usually associated with audio, image, graph, text etc. that are embedded within the Video – these are the information that help us understand the content of the video. Video texts can provide a lot of contextual information on the video clips. In general, there are two types of texts embedded inside video namely, scene texts and artificial texts. Scene texts appear naturally in scenes shot by the cameras. Artificial texts are separately added to video frames (normally in Production Studios) to supplement the visual and audio contents (Lienhart, 1996). Since artificial text is purposefully added, it is usually more structured and closely related to context than a scene text. The text data in video frames contain useful information for automatic annotation, indexing and summarization of the visual information. Extraction of the text information involves the following processes – 1. Detection of Text Region 2. Localization of Text Region from the detected region 3. Tracking of Text from Localized Region 4. Extraction of Tracked Text 5. Enhancement of the Extracted Text 6. Recognition of the text from the Enhanced Input 7. Post-processing (language dependant) of Recognized Text