14th Jan 2013
Google can search images
Gosh I didn't know that!
Google has the ability to index images by creating a visual graph of the image so in conjunction with the image text (file name, alt, title and caption) they can rank images based on relevance to the search.
OK I knew that it was important to name each image file using accurate keywords. And I knew it was important to put these keywords in the alt, title and caption descriptors for each image. An image name made up of numbers is meaningless; to both search engines and users. Yeah, I knew that.
But I found a study by experts from Beijing in April 2008 that blows me away! That's nearly 5 years ago! They wrote a paper called “Page Rank for Product Image Search” which shows algorithms that use an image graph of each image to check similarity between images. They created their own links between related images, allowing a ranking of images according to this similarity. They developed ways to remove irrelevant images so they could automatically show only the best images for that search topic. They included research on user opinion to see if the images fitted the search topic. They created a simple mechanism that enabled them to outperform Google (in 2008) by ranking images as more appropriate and relevant to the search in a way that removes the irrelevant images. They call it Visual Rank.
Gosh, that sounds really helpful.
Precision image search
According to The New York Times, Google created a Prototype for Precision Image Search. But the difficulties that needed to be addressed include image orientation, colour and shading, partly hidden, along with the practical difficulty of analysing every image – there are billions of them!
Search engines rank images based on their local descriptors (the text associated with the image: file name, alt, title and caption). But Google took this further as they have the computing power to analyse large quantities of images and compare them. They have developed ways to tell if there is a face in a photo. Comparing the text descriptors for a photo they can see there are similar image graphs for other images of the same face. They can analyse a photo to see that really is a red car called a Toyota Prius and therefore give you the best images for this kind of search.
Google is learning to "read" images
Since 2008 they have been teaching Google to see images and creating a Google image labeller. There was work done a long time ago into using image recognition as passwords. (People forget random letters, but easily recognize images accurately). Google USA has a mobile phone application called Goggles where you take a photo of something and it will recognize the place or product and give you information about it like price, local sellers, reviews, recommendations… Google is already using face recognition with their Google Plus network and they have called it “Find my Face”. Picassa is using this technology too. Is Facebook?
As at 31st May 2012, according to Search Engine Land, Google Products search changed to Google Shopping. This hasn't reached NZ yet as it’s only US based at this stage. But this image recognition feature that Google is developing to ‘see’ images is a big part of Google’s collection of products to show on their shopping searches. It’s called Snap and Search.
Optimisation of images is needed
So that means more than big brother watching us. He can understand what we share and maximize the speed and accuracy of searches according to images as well as text. With Google’s ownership of YouTube it’s exciting that animations and movies may one day be analysed to improve the internet search experience.
I guess for SEO this means not only do we need to get our alt, title, caption and file names right for images, but we need to have images that are as similar as we can get to the ones that Google Images already show for that keyword search, without cut and pasting the exact same picture as elsewhere. Tricky... But that must be why SEOmoz recommends analysing visual link structures. Yes – another important optimization feature to check…
Image search ranking - Gosh! I didn't know that!
CommentsHave your say …
Have your say