I learned from this site that reCAPTCHA is utilized as a method to digitize distorted characters by offer an unknown one and a known one.
Much like a Mechanical Turk application, ReCaptcha uses humans to translate images of scanned words that a computer couldn’t understand.
And from this paper:
helping to digitize old printed material by asking users to decipher scanned words from books that computerized optical character recognition failed to recognize
I read the paper and found how they estimate the uncertainty:
We start with an image of a scanned page. Two different OCR programs analyze the image; their respective outputs are then aligned with each other by standard string matching algorithms and compared to each other and to an English dictionary. Any word that is deciphered differently by both OCR programs or that is not in the English dictionary is marked as “suspicious.”
The key point here is the two different OCR programs and a dictionary, but, in the scenario of Geo-reCAPTCHA, there is no dictionary to check if an object is recognized correctly or not and hence the risk of missing out on the suspicious images would increase.
Do they outsourced all the images or only selected the suspicious ones? If they did some selection, how did they do that to ensure the true suspicious are chosen and the obvious are filtered out?