How image search works
Posted by admin on April 19, 2012 – 5:39 pm
How image search works
by JV Valdez
Images online can be searched provided they are properly tagged and indexed. Search engines and specialized image search engines do this task of sifting through millions of items online in blazing speeds. There is a widespread misconception that image search is performed based on determining information on the image itself. Currently, however, searching for images is not the same as “watching” the images. Just like websites, image search works the same way as search engines. A user just types in a keyword for the desired picture, then a huge database of indexed metadata is searched when a query is made, the search engine reviews the index and queries get matched with the stored metadata. Metadata is one of the bases for image search but it rarely uses the data automatically produced when a picture is taken by a digital camera. Rather it is usually the description and caption that an uploader tags to an image.
Metadata can serve two functions for online images: proper tagging and indexing, and search engine optimization (SEO). For this reason, SEO comes into play when an uploader wants their image to be searched more. An image is properly tagged so it avoids clutter when grouped with numerous other images and aids the website in gaining more traffic. Appropriate and highly-searched keywords and search phrases are employed to achieve this goal.
The big three search engines, Google, Yahoo and Bing all have facilities for image search. A separate tab can be found on their websites where users key in search words. Results are given through thumbnails so that images can be reviewed right away and sorted according to relevance, file size and rating. While major search engines provide image search and retrieval services, they do not host the images themselves. Specialized image sites such as Flickr, Fotopedia, Open Clip Art, Pbase and Getty Images host images and run image search engines for free or for a fee.
Then there are novel ways on image search. TinEye can help a user pinpoint the location of a particular image used on the web. A reverse image search is performed when an image is uploaded where the website creates a compact and unique digital “fingerprint”. It then compares this fingerprint with other signatures from a large database in their index.
A true form of image search, which is essentially “viewing” the image for its contents, is content-based image retrieval (CBIR). Also known as content-based visual information retrieval (CBVIR) or query by image content (QBIC), image search is done by analyzing the actual contents of the image instead of relying on metadata. Current techniques in achieving CBIR is still in their infancy, but the basics has been properly laid out. Image content refers to textures, colors and shapes, vital image features to detect what the content actually is. Having humans to manually enter keywords for every feature on each image is expensive resources and time-wise.
In November 2011, Microsoft purchased VideoSurf, a California-based provider of technology that permits users to conduct visual searches on videos, for $100 million. Although it focuses mainly on video search, it definitely paves the way for a better and more efficient true image search. That amount of money is already significant among tech circles, but lays the foundation for a more dynamic image search and true online experience.