What if Google is looking at the nature of the image as opposed to reading it? We know Google is good with images when identifying an image being used elsewhere and offering similar images. Therefore, maybe the image scanning is a topic-based scan and not an OCR-type scan. We check this out further.
Test Downloads
Transcription Below
This is a follow up. This is actually another Josh Bachynsky inspired one where Google have acquired a couple of image recognition companies specializing in reading images and understanding them. In particular, the technology that was purchased could read text on an image.
Last month we tested whether we could put text on an image and run it through AdWords for their keyword suggestion and kick back the suggested text, the text that was on the particular image.
After we ran the test, there were some thought on maybe it cannot read the text but it might actually see the image and understand what the image is. So we rerun the test with no text in the image but just the actual image and see if we could get the image planner to kick back suggested terms based on those images alone.
The images that were used in this test were images that were already used by Google so if you were to search for a particular word, it was the image that came first in the image search. It is an image that Google knows and understands, we put that on the page, we run it through AdWords for the keyword planner and the suggested terms.
In this, we got the football match, will Google kick back terms related to soccer or football? The next one was online accounting courses. These were the first things that came up. Will it kick back suggested terms related to these images?
In the end, it couldn’t do it. The suggested results had nothing to do with what the images actually were. So, Google is not reading an image – just an image itself – with the technology that Google has. Right now, they are not reading and understanding what that image is and using it in terms of understanding and indexing that page.