Image search VS Visual Search

April 29, 2008 by pmelcher.

Google is thinking about changing its Image search algorithm. Currently it has a convoluted way to return results. As you probably already know, its a basic “text” search which looks at the file name, “alt” comments and words around the image to declare an image a valid candidate to a search. Meaning that if you search for “cat” for example and someone has named an image of a truck “cat.jpg”, has put the description in the “alt” comments as “cat, cat and more cat” all that around an article about how great his cat is, then that image of the truck will appear in your search.

Not very efficient, is it ? And as previously written here, the image could be completly out of focus and grainy, as long as it meets all the requirement for Pagerank, it will appear high up in the results.

Called “PageRank for Product Image Search” and presented at the International World Wide Web Conference in Beijing by two Google staff engineers, it is aim at becoming the new VisualRank.

Claiming to be an image recognition system and using advance object recognition, here is what it does. It scans all images and looks for patterns, regardless of what object is in the image. After a while, it will see that some images or at least part of the image have the same pattern . Those will be linked. The image or images that have the most similarities with all the others will be pushed to the surface.It gives you a result like that :

See that image in the middle ? It contains all the attributes of the others, thus its the most relevant.

If applied this will create a headache for the photo business. Since this search is really made to search for products to purchase and not for images to license, it is counter productive for our industry. It will not return the best image, the most liked or the most striking, only the most banal, the most common. Ouch !

It will favor non exclusive images, think RF and microstock, over RM images. It will enhance the most used images not the best ones. It will slowly bring IPTC kewording to obsolescence.

In order to bring traffic to its website, a photo agency or photographer will have to post  images as much as possible everywhere all the time.The same image. Thus an image with a lot of various usage will be the star, while news images, who usually have a shorter life-span, will not score well. But an image of a spoon might become a superstar. Especially if it is sold everywhere

Google hates photography. Or rather it sees  it as a tool, not as an art. Another way to index the world.It will become harder to find great images with Google and that will continue to open a door wide open for anyone looking to create a search engine for photography with a ranking system based on quality and relevancy. An image search and not a visual search.

More on the emmerging proposal at Techcrunch.

Posted in IPTC, Search, technology, keyword, google, filter, flickr, editorial | Print | No Comments »