New Google Tool Cuts Down Irrelevant Image Search Results

Dennis Faas's picture

Google now has a secret weapon in the ongoing fight against irrelevant image results: VisualRank. VisualRank is basically an algorithm that blends image-recognition software methods with techniques for weighing and ranking images that look similar.

You are sitting in front of a computer screen, relieved to have finished a lengthy term paper on wild animals. You decide that adding a picture of a jaguar would be a nifty way to give your title page some flare. You connect to an image search engine and type "jaguar" in the designated box. Suddenly, your screen fills with images of automobiles, rock bands and shoes until you eventually stumble upon a distorted image of the animal you were looking for in the first place.

Sound familiar?

An image search engine functions in a very complex manner. The computer matches word(s) typed in the search engine with "hit words" found on random web pages dispersed all over the Internet. This has led to a number of misplaced hit words resulting in many irrelevant images.

Even describing the image in great detail sometimes leads to incorrect results. Typing "jaguar animal" could reveal images of "Uncle Frank Riding a Jaguar" or the entrance to the San Diego Zoo.

Analysts believe that Google is attempting to conquer an impossible feat: cataloguing the billions of images found in the "Google Images" search engine queue.  (Source: nytimes.com)

In an effort to make the effort a bit more manageable, Google announced plans to focus on recognizing only the 2000 most popular images first. These include major mainstream products like the Apple iPod and Sony PlayStation.

VisualRank uses a scoring system to check the relevance of an image. The software has already been able to filter out 83 percent of the previous irrelevant image results.

Riya, a small Internet start-up company, developed Like.com in 2006. The website offers a service in which online consumers can select a particular visual aspect of the products they wish to purchase. The service is unique because the software is able to match shapes and objects with images instead of the text-based "hit words" system currently being used by many leading image search engines. (Source: aviransplace.com)

While not all images are expected to be refined, Google does plan to continue their promise of offering "the most comprehensive image search on the web". Whether image search engines will eventually become irrelevance-free is something Internet users are waiting to see.

Rate this article: 
No votes yet