Main » 2011 » Март » 16 » Character recognition

---

11:03 Character recognition

Here is the recognition process. In this paper, I used python, but nothing prevents to transfer the project to other languages. Manually creating a database with ready captcha (about 100 pieces). For each character a couple of audio captures, in which it occurs only once, and all other characters are unique, ie in different captcha is not repeated. For example, for figures - 2, take these: AS2DE, 2ZTKJ. In the chosen captcha, the usual exhaustive search looking for the same longest match sequence. At the output we get about 2000 characters. Controlling, lest we caught a fragment of 'pause'. Add the result to the database. Example of simple search for the two captchas: defcompare (letter, filename1, filename2): tfile1 = ; filename1 + '. mp3' tfile2 = filename2 + '. mp3' f = open (tfile1, "r") test1 = f.read () f2 = open (tfile2, "r") test2 = f2.read () cnt = i = j =- 1 ; k = 3000 for item in test1 [:-k]: i = i + 1 ; j = i + k cnt = test2.find (test1 [i: j]) if cnt> 0: res = test2 [cnt: cnt + k] ; f3 = open ('sources /' + letter, 'w') ; f3.write (res) return return On That's all, the recognition result 100%. Now, when the robot sends our news on digg.com, he finds the address on page image Captcha, replaces to mp3, voice acting queries by using cookies, we find the desired 6 characters, comparing with its own database and sends the result. All the news from your site will be posted on digg.com a few seconds. If your site has audio captcha, I recommend to reject them or to secure the following recommendations: Use of different people with different accents and intonations. To vary the sound level, especially in the background. Add noise generated at random. To crack a protected audio Catcha, a simple method of comparing pieces of mp3 files can not give positive results. In this case, it is recommended to use special filters to process audio tracks and noise removal. After that, alternatively, you can apply neural networks to analyze the sequences. Of course the result will be less than 100%, but will remain on the level. Also you can try the services of speech recognition. The best I've seen - it is Google Voice, it is only necessary to send voice mail on our mp3's and some time to get transcription (it would be interesting to see the results).

1 2 3 4 5 Views: 415 | Added by: w1zard | Rating: 0.0/0

Total comments: 0