Don't even think about it.
"Similarities" in an image are complicated enough: a shade off, a slight change in size or alignment, a small rotation - these are difficult enough for a computer to spot in an actual picture: image similarity python - Google Search
But converted to text? once you do that, you have no real idea even what shape
the image was, much less what format the data might have been (and a JPG file content will be very different from a Bitmap or PNG file!).
Follow a few of those links, and use the packages to compare actual images, not a text "representation".