Blocking most images would possible if it was necessary and if Google provided a simple API or the source/database of Reverse Image Search, which can generate a few short statements to describe an image. The rest would be as simple as querying a small database of terms and checking with fuzzy search if they were in the description.