I'm not sure they call it delta image in literature, but here's a brief explanation of it:

An image is essentially an array of pixels, each pixel is a vector of numbers that represent the Red/Blue/Green (RBG) components (or any other representation format).
I have the current image that you uploaded without the characters, when you upload the new image, I'll download it and now I have 2 images, subtract the values in each pixel in image 1 from the values of the corresponding pixel in image 2, now you have an image (array of pixels) that contains the difference between the 2 images, so you can see exactly which pixels were modified.
Pixels with low modified values (that can't be distinguished by human eye) can be zeroed out to reduce the "noise" of the image, this results in an image with only pixels that have significant change between the 2 images, all that's needed is to run some algorithm to detect those small characters locations (group of neighbor pixels)