In the last topic we mentioned some notable video retrieval systems developed by research groups, some of which have led to larger deployments on the Internet which attempt to index significant amounts of published media. We noted SpeechBot as an early technical success in this domain; applying speech recognition for retrieval of large repositories of Web media. More recently Podzinger based on BBN’s highly regarded speech technology began indexing podcasts based on speech content and RSS metadata. The site indexed both audio and video podcasts and provided extracted context for query results. Rebranded “EveryZing”, the site has this to say regarding ingestion formats: “EveryZing can index, search and reference English and Spanish media (language tag in RSS file must begin with “en” or “es”), that are formatted for audio as MP3 files or for video that are formatted as mp4, mov, m4v, flv, mpg, or mpeg files.” [EveryZ-ing07].
BlinkxTV gained attention using speech recognition and other methods such as link context derived from page layout proximity to improve retrieval accuracy. They also have broad coverage of many IP media sources. Additional features such as displaying results in a grid with animation also give Blinkx extra panache and may prove to be valuable for results browsing. Based on technology from Cambridge University, the site reports that 111 patents protect the technology which was developed over 12 years, and that there are over 18M hours of index video [Blinkx07]. In addition to speech processing, any other available metadata, including closed captioning, is used to build the index.
There are hundreds of popular video sharing sites, and of course Google’s YouTube is the most well known. (Google has its own video site developed prior to its purchase of YouTube.) With the advent of Podcasting, many sites have sprung up that are focused on indexing syndicated content feeds since the technical barriers to entry are relatively low. These sites enjoy the more organized metadata extracted from RSS or similar syndication formats and the technical work is largely in XML processing, database optimization, UI generation and scaling to handle load. The popularity of various sites is quite dynamic, and many social networking sites are participating in video search both as a content source and as a fabric for posting commentary on clips pasted into user pages. In addition to YouTube, sites such as Dabble, Truevo, Clipblast, Metacafe appear on most lists of popular video sites. In addition to Google, the other established names in the search or portal arena such as Yahoo, AOL, and Microsoft (Live Search, Soapbox) all have video search strategies, some involving partnerships. Most have multiple activities underway focused on sharing and mobile as well as traditional video search.
Conclusion
Digital video technology has been steadily improving for many years, and recently the cost reductions combined with ease of use for capture and editing has resulted in rapid growth in the amount of video being produced. User generated and enterprise promotional material have mushroomed and augmented traditional video production sources. Distribution to mobile devices and set-tops via IPTV, unheard of a few years ago, is commonplace today. More and more video material, even high quality material, is published on the Web – particularly as more people turn to their laptops as an alternative to watching TV. Further, the Web is extending to an ever increasing array of mobile devices. Therefore Web based video search will be a key enabling technology going forward. Media processing technologies, in addition to metadata handling systems, are available today to help users locate desired content and to navigate through huge amounts of video material for both educational and entertainment applications. These technologies, however, have significant room for improvement through algorithmic innovation as well as through the application of novel engineering techniques to improve the overall efficiency to make indexing large video collections practical. These conditions provide great opportunities for research and development to have a significant impact on people’s day to day interaction with video information.
