11:03 Structure captcha | |||
And then we stop ... Habre already ran the ways to recognize characters in the picture with the aid of python, OCR and neural networks. The most fully this topic are covered in the article distinguished Indalo. But this method does not give 100% probability of detection and relatively complicated to implement. Knowing that there is always another way to solve the problem easier, I accidentally saw an interesting phrase: «Can't read the text? Listen it ». Listen, I've noticed that all the characters voiced by a speaker and always the same, without the noise and extraneous sounds. And indeed scoring is designed to help people who are not able to discern all the letters, enter the correct characters. If this method is easier for human perception, it accordingly should be easier for the bot. When zapolenenii Forms site gives us this kind of picture (Warning requires cookies!): Http://digg.com/captcha/2c7ea3845d5ddfc5a7461c5429b6a7e5.jpg The sound file will look like this (Warning requires cookies !): http://digg.com/captcha/2c7ea3845d5ddfc5a7461c5429b6a7e5.mp3 After these experiments, we found out that a fragment of each letter is about 2000 bytes. In the background noises are present, but they do not sgenirirovanny randomly, and the same character at different captcha absolutely identical. Therefore, our mp3 files should be viewed as a simple array of characters to search for these fragments. | |||
|
| |||
| Total comments: 0 | |