One of the biggest stumbling blocks for many small business sites (well, any site, really) when it comes to search engine optimization is the lack of text on their web sites. After all, it's well known among SEO's that search engines cannot read the text in an image. Or can they? Bill Slawski walks us through a few new Google patent filings today that may point to a change in this long standing rule of SEO.

Bill writes:

How easy or difficult is it for a search engine to recognize text within digital images and video, and index that text?

Three new Google patent applications explore that topic, and describe some ways in which Google might try to capture information from text within images.

Capturing Text from Street View Images

This patent filings don’t address text found within headings and logos, but rather much more complex pictures, including street scenes of the kind that might be taken for instance, when filming streets for something like Google’s Street Views (video).

While much of the patent talk focuses on identifying the text within photographs taken of street fronts and shops for services like Google Street Views, it isn't a big stretch to think the engine might be able to port this technology over to graphics on a web site.

If you look at it from the perspective of my Pinocchio Effect theory, it makes perfect sense.

If you're not familiar with the Pinocchio Effect, here's the general idea:

You see, deep down, search engines want nothing more than to be real boys (or girls). That's right, it's that simple. As search engine engineers gain more and more ability to tailor the algorithms, their ultimate goal is to help the search engines make choices the way that people do.

One of the primary things human beings can do when looking at and valuing the content of a web site is to read the text made up of graphic files. I can visit an all graphic or all Flash site and read the content just fine. (Assuming I'm using a browser that will display the images or Flash.) Search engines have long been unable to read any text contained within an image.

If people can read the text and make judgements based off of it, search engine algorithms will naturally be looking to do the same.

According to Bill's post, it looks like Google may be well on the way to getting the technology in place to pull this off.

It's important to note that nearly every example in Bill's post is based more on being able to tie text to photos of store fronts and such than of graphic text blocks on web sites. That means chances are good that we'll see this coming into play in areas like Google Maps, Google Street Views and Image Search before we see Google attempting to actually index any content contained in the graphic text blocks on a web site.

That means you need to continue to make sure you're offering up your site content in a format that both humans and search engines can read. It also means you might want to run outside and make sure the address numbers or your business name's sign aren't crooked.

January 4, 2008

Jennifer Laycock is the Editor of Search Engine Guide, the Social Media Faculty Chair for MarketMotive and offers small business social media strategy & consulting. Jennifer enjoys the challenge of finding unique and creative ways to connect with consumers without spending a fortune in marketing dollars. Though she now prefers to work with small businesses, Jennifer’s clients have included companies like Verizon, American Greetings and Highlights for Children.


Hi Jennifer,

The patent filings do focus upon text identified and extracted from photos, like those found in the Street Views project.

But some of the patent applications on Book Search from a couple of years ago also describe using Optical Character Recognition to extract text from images of books, and to distinquish between different structural elements within books. That way, the search engine can tell if something is the book title, or a chapter title, instead of body text.

At least one of those patent filings hints at the same type of technology being used on the Web. Some interesting implications from that. :)

It's difficult to tell how much of that might be done on text within images on the Web these days, but I expect that it could benefit a number of sites to have Google understand text within their pictures. Should be interesting.



Thanks for the added insight Bill!

It makes sense to see the focus on things like Street View. There's obviously a good secondary information source for matching up the information. For instance if you KNOW the storefront is at 301 Fifth Ave and that's what text is appearing in the photo, you have confirmation. Same for store name and such.

In terms of the books, also makes sense there as we know there are already software solutions that will "read" a scanned page and convert it to text with varying levels of accuracy. Straight text in a standard font would seem pretty "doable" but as you get into more complex graphics like often show up in some site designs, I could see how it would get more complicated.

Thanks for keeping us all updated on the progress here!

You're welcome, Jennifer.

Great point, here:

For instance if you KNOW the storefront is at 301 Fifth Ave and that's what text is appearing in the photo, you have confirmation. Same for store name and such.

The authors of the image/text patent filings do actually mention that they can and will look at supplemental data, like telecom information when they know a location, to help them in their efforts. They also mention associating images with GPS information, and other location data, when taking pictures. I'd imagine every bit helps.

The book scanning process has to be painful in someways. In addition to trying to address many possible different fonts, they are trying to make it a global effort, so many different languages are also covered.

But you're correct. Complex imagery in site designs could create some additional problems that could make indexing that kind of image text pretty difficult, too.


Comments closed after 30 days to combat spam.

Search Engine Guide > Jennifer Laycock > What if Search Engines Could Read Images?