With Broadband, content explosion, increasing use of the internet for day-to-day activities, using search technologies that were developed for searching flat HTML files would not be sufficient to meet current day requirements. Special technologies are needed to search multimedia files, provide non linear search capabilities, find patterns and provide search results by factoring multidimensional attributes are the focal area of the search industry for now and the near future.
WIRED magazine has come with an excellent article about the research and advances being made in the realm of video search. Excerpts with edits:
Searching the internet for images or videos has always been a tough proposition. Researchers are developing visualization technologies that can "see" inside images, reducing search engines reliance on text-based image tags that are easily manipulated. Search companies testing the waters of image retrieval see it as a lucrative method of connecting advertisers with customers, according to Chris Sherman, editor of the SearchDay newsletter Yahoo, Google and MSN,as well as AOL's Singingfish have launched websites for searching static images and video.
These websites reliance on searching text descriptions of images significantly limits their ability to deliver the correct pictures, according to Sherman. Very few content creators take the time to add text descriptions (metadata) to visual content while it is being created,so many images can be missed by search engines.Websites looking to increase their traffic frequently mislabel their images. "People often use metadata (such as claiming to have pictures of XYZ) to lie and promote their sites," he said. Pornography sites and websites known to deliver spam often claim to have pictures of celebrities or even popular children's toys to trick surfers into visiting them.
A group of European researchers are developing technology that can recognize everyday objects in digital images. The image-processing software looks for "key patches" in an image to determine the relative positions of different shapes, such as tires and a car body, or a beach and ocean waves, to categorize the image's contents.The software has learned hundreds of objects since development began in 2002, and can be used to categorize images and automatically create image tags. The software can look for images similar to those it has already scanned and "knows,". The software is currently being tested on a variety of images, and the researchers continue to add new object categories. Companies such as clothing stores or sporting goods companies would jump at the chance to have a Google image-search result in pictures displayed with their products.
IBM's Pervasive Media Management group is developing visualization software that can identify objects contained within one of the web's fastest-growing content categories - video streams. The software identifies groups of objects within a frame to form concepts that can be easily searched, such as an airplane with a cloud and sky backdrop that would be categorized as travel.Categorizing the content of video through human labor can take 10 times as long as the duration of the content, as per IBM. The software can be trained to recognize images by providing it with a group of similar images.IBM is working with broadcasters CNN and ABC to identify concepts that can be used to classify news footage.
Yahoo is trying to improve the quality of video searches by asking video producers to improve the labeling of their content. Yahoo created the Media RSS format, a syntax for labeling video content for its video search site, which debuted in December. Karen Howe, vice president and general manager of Singingfish, said she has reservations about using image-recognition software. "Technologies for image recognition and scene summarization are pioneering areas that we definitely find interesting, though currently they are in the early stages of development and not cost-effective for commercial purposes," Howe said. Indeed an interesting space of action and advances in this area would be the key to making the internet more popular for education and commercial purposes.
|
