Search for "heart" of the video

Text: the Internet weekly reporter Yang- what is the Internet's most popular vocabulary? answer ranged, but "video" must be one of them. With the DV camera phone, video equipment, such as individuals, as well as the popularity of the Internet broadband applications, more and more video content to the Internet. Video search application has been referred to a schedule, but the product on the market today and also can't let people complete satisfaction. How can you achieve precise search? how to quickly understand video results content? the Internet Weekly reporter Yang-by Microsoft Research Asia's first victory of in-depth interviews with researchers, as we unveil a video search in the "heart".

Find ignored the "cat"

Even for a duration of only five minutes of video, or it may contain a lot of content, such as landscape scenes, character dialogue, or even a lying in a corner of the cat.

If this video of the theme with the cat no coherent, it must not be manually edit the video file name and description. That is, to text as an index of video search engine, it is not possible to search into the cat. Of course, a lot of similar details also are "forgotten".

Compared with the text and pictures, video, sounds and texts as a collection of time series, on search technology requirements are higher.

In the current market, most of the video search engine used by almost all the text index. So, is there a better way, let us be able to search for more details, researcher, Microsoft Research Asia?, first win introduces a truly "search inside the video to" index.

This is based on the index of the "content", will first show the search engine "recognize" the image of a cat and his "translation" as the text through indexing.

In this way, when a video file in the cat, it will be automatically recognized. Similarly, you want to search for a sports competition in a link, or you can easily implement. For example, if you want to search for a baseball game in the "pitch" (pitchview) action, namely the "ball from pitching area to batter" scene, usually video text will not be careful to describe the action, but this content index you can accurately.

Content index as an effective complement to the text index, you can make the search engines to achieve a more precise query.

While these actions have a premise that the need to set some of the training data, allowing search engines to "learn", that is, to let his "know" what scene is the "pitch". According to Microsoft's experts, the future will be completely divorced from the training data, let the search engine directly through the network for intelligent learning.

Clip-rendering

When you get search engine listed a bunch of random results, another problem appears: "which is what I want?" in achieving an accurate search, the future of video search engine to do another thing good rendering — to let users quickly understand a video full of content.

In reality, you use the video search often encounter such a situation, when a search is "football (Soccer)," but you want some wonderful game footage, the system may also lists is Stephen Chow's the co-writer, even "Beckham" entertainment video news.

In response, Microsoft Research Asia in the search results for category segmentation functionality, you can follow the duration (duration), type (genre), format (format) and sorting criteria for filtering. Each standard includes several segments of the category, for example, select the "type" is selected, you can follow the "sports" and "film" and "news" category breakdown. Thus, the messy results becomes clear.

As Internet

For many video is based on the HTTP protocol, which means that each play a video, download, play again from the beginning, that is, you can only play in the order, the inability to achieve the jump. The Microsoft Research Asia is currently a technology can implement HTTP free jump.

According to China to win researcher, video rendering in the key point is the current video summary technology, Microsoft Research Asia total research out of the 5 types of rendering video search results, not only can be dynamically navigate to the keywords are located then a keyframe, and will keep you in the pages of 10 to 30 seconds of playback preview.

Where there is a rendering of the list, in addition to providing a "dynamic thumbnails", like DVD browse function before sowing, each video is divided into independent play five chapters. In this way, users can get a sense of the content of the video file. While the use of this technology provides scalable rendering, you can achieve a personalized piece breakdown, for example, when you press the "x 4" mode rendering, the original video is divided into 32 clips.

This is not all.

The future of search engines also joined the Web2.0 elements — users can on a video in a (few) clips, create personalized comment tags (Tag), and may collect and share. This feature will greatly improve existing video sharing community, user experience, increase user stickiness.

Improved version of video advertising

"Johnson Johnson after three pedestrian, advertising.

"The audience is always interesting to know, that after you finish this sentence, what douwentao and guests talk something — TV is in strict accordance with the time spots, that is, when the program is broadcast to a fixed point in time, spots, but its content is intended for all users, not necessarily directly related with the program.

Internet video advertising provides a more flexible, individualized criteria.

As the main means of profit, it will have a direct impact on the development of online video. But due to technological development, the current market products also can do a video on the head, tail position insert advertising, cannot do with the video content match, smart insert.

To do this, Microsoft Research Asia is a combined search of video advertising platform, you can break through the existing time insert advertising model, from the user experience and advertisers benefit two perspective, real time video content in the right place at the right spots.

In addition, the video ads in the surrounding text and replace is also the scene needs further study.