Why Comprehensive and Relevant Image Search Matters

A Web surfer is not the only interested party who wish to get high quality search results when looking for an image.

Search engine companies are seeking the ways to improve image search. The more relevant content found, the more precise binding of ads to content can be performed. However, finding a relevant image is not an easy task because currently search engines can operate with very limited information about images located in the Web.

In our days, the Internet is not only about Web pages. The best answer when a user performs a search, as Sergey Brin, the Google co-founder, noted on a conference call outlining third-quarter results, is not always a Web page. Although Brin applied it to video, it is obvious that an ability to find the image with content a user needs is the challenge that search engines have to meet.

Better search results mean better ad targeting. Binding ads to search results is what makes search engine giants worry. Yahoo! reported that they are not satisfied with their third quarter financial performance. “We continued to grow and believe that we outperformed the graphical market but not at a rate that met our expectations,” said Terry Semel, chairman and CEO of Yahoo!

Disappointing results were caused by low sales of online ads for cars and financial services. Analysts say that Yahoo! still fails in precise tailoring of search results to ads and therefore do not meet advertisers' expectations. In Yahoo!, they understand it very well and are looking for new ways to find more content and improve context advertising. Yahoo!'s efforts to integrate search results with images from Flickr is an excellent proof.

This is when the real problem comes. While it is not difficult to index images that are already tagged with Flickr tools, pictures that are not shared through Flickr (it is pretty obvious to assume that their number is incomparably more) cannot be achieved as effectively as Flickr photos.

Google has developed its own way to tag images placed somewhere in the Web and not necessarily in Google-owned Picasa. In September 2006, Google released the Google Image Labeler game (http://images.google.com/imagelabeler/) that should help the search engine improve its search results through quality tagging. During the game, a user is randomly paired with a partner who is another Web surfer. Over a 90-second period, both users are shown the same set of images and asked to provide as many labels as possible to describe each image you see. For example, when seeing an image that shows the Real Madrid football team, a user can offer labels “football”, “Beckham”, “players” and so on. When the label offered by one user matches partner's label, the user earn some points and move on to the next image until the game is over.

Luis von Ahn, the Carnegie Mellon Institute professor, who is standing behind the Google game concept, thinks that in two months, all images on Google Images could be labeled. As a result, it is expected that image search will become more precise.

Although this is the great idea and good implementation, the main problem is that the minimum unit of content is still image as a whole rather than individual element of the image content. If you succeed to find the group photo of Manchester United by using “Beckham” as a search word, how do you know which of them is Beckham? While a traditional text search enables highlighting the word you are looking for, there is no way to highlight the individual object that matches the search request unless a tag which clearly and unambiguously refers to this individual object is assigned.

An ability to work with images on a more granular level would provide much more opportunities both for search engines, which are interested to find as many relevant search results as possible, and image owners who are interested to bring more relevant traffic to their Websites.

For search engines it is about finding more relevant image content and referring to found objects more precisely. In terms of revenue, it means finding more specific results that can be used for more precise ads targeting. The more specific objects are tagged on an image, the more opportunities for advertisers can be offered. For example, a photo that shows Mercedes and labeled with tags like “Mercedes”, “car”, and “auto” can be associated with advertising of a car dealer. However, if individual elements, such as tires or car alarm system, are tagged and highlighted depending on the search request, ads of manufacturers of tires and alarm systems can be also shown.

For image owners it means making images easily reachable by search engines, increasing ranking level, and therefore attracting more relevant traffic. Object-specific tagging would take a few minutes that would be even less and easier than developing metatags for search engine optimization. Moreover, an ability to define individual elements on images provide more opportunities for SEO because object-specific tagging allows Website owners to go beyond traditional meta descriptions and put more information into a single image.

Although image labeling proposed by Google is a great step forward, the further direction should be defined right away. Image labeling should now go down to a more granular level and let search engines work with picture objects rather than with images as a whole.

The question is what exactly should be done. See how Cogitum LC solves the problem of labeling individual elements of an image.

Send Us Your Feedback

We are always happy to get your feedback on the articles. It helps us constantly improve the content and provide you with the most relevant information.

If you know or want to know something that covered or should be mentioned in our articles, please feel free to contact us. We will be glad to get to know your opinion.