python - Image Similarity that supports Resizing and Cropping

Question

I've read a lot about image similarity and comparison algorithms (focusing on Python), and I tested some of them out, however I haven't come across the correct algorithm (or combination of indices) for what I'm looking for.

Let's say I have a base image - a screenshot of webpage X from my browser. If a different user takes a screenshot of the same webpage on their PC, the same page can look different in a two main ways:

• Texts/images on the page appear bigger/smaller, and even located a bit differently (because of responsive design for different screen sizes)
• The user might have scrolled up/down a bit before taking the screenshot.

I'm looking for the best algorithm/method to make sure I get a high similarity score for these cases (and of course as low as possible for when it's a different webpage). For example, I tried using the SSIM algorithm, which gave a high similarity score for these cases, however it also did so for very different webpages. Any ideas?

EDIT: Another idea - how about specific object recognition? i.e. choose a few objects that appear on a webpage, and look for them. Would it be possible to find them even if they appear in different locations and/or sizes?

question from:https://stackoverflow.com/questions/65651434/image-similarity-that-supports-resizing-and-cropping

Answer 1

Waitting for answers