The plan is to have a weekly journal on an idea or a product, that I understand or learnt that week.
Let's get started and this week's cynosure is "reCAPTCHA"
We, often, see sites that ask us to enter the content of a distorted or skewed image of letters or numbers, that is known as CAPTCHA.
The widely known fact is that it enables the site to distinguish a human from any automated bots or scripts.
It is so reliable, that vast number of sites are using it, and about 200 million captchas are answered by humans in a day. Roughly it takes about 10 seconds per person to answer a captcha. In summation, each day more than 150,000 hours of human effort is consumed by these Captchas, which does nothing more than confirming that the detail is entered by a human.
Could this human effort be used for a higher purpose? Yes. The answer is reCAPTCHA.
reCAPTCHA uses this humongous human effort to digitize books.
At present, OCR technology, that is used to extract content from scanned material including books, is not perfect and the success ratio is only about 60 percent. reCAPTCHA is used to fill the gap by making the humans to identify and correct the word.
Here is how it works.
- reCAPTCHA shows two words to the user. One that is known to the computer and the other one that the computer is not sure about.
- The computer authenticates the user based on input for the word that it knows correct.
- The computer records the user input for the word that it is not sure about
Same reCAPTCHA is shown to multiple users and their input for the word, that computer is not sure about, is aggregated and the most common occurrence is the suitable match.
This method is found to be producing very accurate results, where the success rate is about 99.5% and has been used to digitize the old editions of New York Times and books from Google books.
In India, there is an active research on OCR for digitizing old transcripts in Indian Languages. Having reCAPTCHA in Indian Languages will help significantly for these projects. Thanks to Google transliteration, typing in Indian Languages is no more an irksome job. Also, choice could be given to the user to choose the conversant language.
So, every time a reCAPTCHA is solved, we can be glad that we augment the human knowledge by synergy.
Interesting TED video about reCAPTCHA: