Innovations on Image Search
by
JV Valdez
Unless you are a hard-core image scavenger, there is a great likelihood that you have never heard of Google’s Reverse Image Search. Despite the giant search engine’s ubiquity, this corner of Google formally called “Search by Image” still waits to be discovered by the common web user. What it does is it looks up for the origin or URL of a particular image. So let’s say you receive an intriguing photo of a person you do not know, just upload (or drag) the picture to the website and a list of results pops up instantly. It is quite a nifty tool, and to an extent, an addictive chore for first time users.
Reverse image search involves analysis of the image, where a mathematical equation is created based on lines, colors, proportions, shapes and other factors. A match is then made against images already hosted in Google. Since the method is somewhat of a prototype, results also show similar images to direct users to other possible matches. Text results that describe the image are shown as well, particularly during instances when Google is quite sure of the result.
The launch of this service in June 2011 by Google posed a lopsided but nasty competition against perennial reverse image search engine TinEye and a host of other low-keyed image search outfits. Many experts thought that Google, being the behemoth that it is, would just buy out pioneering TinEye (it started May 2008). Instead, it decided to develop its in-house service ready to gobble up competition. In a way, this is good, as more players in the market mean more innovation. In its inaugural, Search by Image had some serious lapses as it almost always failed to perform facial recognition. The expert on this is TinEye, but Google has been catching up in this department. Other noteworthy reverse image search engines lurking in the background are Byo, Gazpopa, Idee, SnapTell and ImgSeek.
Facial recognition on images has been a thorny issue on privacy because, obviously, anyone who has their photo online can be traced. Facebook, and now Google, has always been in the hot seat during such discussions. Ethical issues aside, reverse image search is a big leap from the last decade’s image search services. What many people do not know is that normal image search works just like your average search engine. There is no intricate algorithm employed by search engines when a user looks up for an image. Such process is actually ruled by keywords and SEO. It starts with the file name of an image, followed by description in the metadata. While all pictures taken by digital cameras and scanners are automatically assigned metadata, these are not the data deemed important for image search. An uploader can modify the metadata for the purpose of proper indexing and tagging and search engine optimization. In other words, for a search that is supposed to be visual, text is king.
An uploader can type in descriptions such as “LA Lakers” or “Jeremy Lin” to an otherwise irrelevant picture of gardening tools. However, search engines are smart enough to discard dubious keywords. Context is the key here–Google, Yahoo and Bing analyzes the whole web page where the image is from and relates the keywords with those found on the website. Appropriate URL of the web page and the image itself becomes a big deal most of the time. A pattern has to be observed, and if there is a match, the search engine will most likely push the image at the front of the results.
Image search engines deliver results clustered according to relevance, and search modification can be done based on subject, image size, color, image types (photo, face, line drawing, clip art), layout and posting date. Bing even recommends related search phrases. These results are thumbnailed and images can be zoomed in when the mouse hovers on them. In the past years, results are paginated, but nowadays, there is a seemingly endless scrolling down for results.
Despite improvements, a true form of image search is still wanting. Text still commands results and essentially “viewing” images are in the prototype stage. Viewing images for their content is called content-based image retrieval (CBIR), content-based visual information retrieval (CBVIR) or query by image content (QBIC). Instead of relying on metadata, analysis of the actual content of an image is performed, so for a picture of a meadow, an algorithm is utilized where a program sifts through billions of images and look for those contain a combination of grass, greens and open land. In essence, a search engine becomes a detector crawling for colors, shapes, textures and shapes.
In November 2011, Microsoft finally took notice of the potential of true image search and purchased VideoSurf, a California-based provider that allows users to conduct visual searches on videos, for $100 million. While the company specializes on video search, its acquisition by a tech giant signalled a more exciting and efficient image search in the future.
JV Valdez writes about technology–its development and innovations, and how people respond to them. He also writes about travel and political affairs.
Video Search
Article Source:
Innovations on Image Search