Does google aggregate&correlate data in vision algo? For example for street numbers they not only have picture of a number, they also have knowledge of all the

Does google aggregate&correlate data in vision algo? For example for street numbers they not only have picture of a number, they also have knowledge of all the other numbers on that street and guesses for those other numbers. Easy to guesstimate order of a number by checking neighbouring ones. Same for book words, they have n-gram database. http://storage.googleapis.com/books/ngrams/books/datasetsv2.html http://storage.googleapis.com/books/ngrams/books/datasetsv2.... Thats a lot of useful MAP/ML data. But the example they give for the new captchas all look like random crap, "mhhfereeeem" and the like. Its like they are not interested in structure, just pure geometry of letters/numbers.