Well not that simple for a machine. The chars are all joined with noize. Though I try my best if you can deliver at least 20 captchas on that type with correct answer in file name.
Yes, however, it always helps and speed things up if you can deliver at least 50+ captchas of that type zipped together (filename of the captchas is correct answer).
the second zip (200 images) I downloaded with Ubot Studio. I little program just do download and save the captchas. My guess is that Ubot Studio converted in PNG, but still kept the jpg extension.
with all those variations with strikeout noize it will lower the success rate. I guess you only sent me the hard / unsolved once. When I train against that set, it is not really reflecting the actual solve rate as the more easy captchas are left out and the balance between easy and hard one is not correct. However, I add yours and optimize based on it.
Naw, I just want to explain why the actual solve rate is bigger than the one you get when just training against the captchas you sent. When you scrape some new samples, it's a good mix between easy and hard to solve captchas.
If you load just the hard once in that didn't get solved, you get of course a way lower success rate which is not really reflecting the REAL success rate.
Sven, Here goes 130 new captchas with answers for nfp.fazenda.gov.sp.br. Most of them are right, but a few are really difficult even for a human being.
Yep. but still... the most of the captchas have this rainbow colored, but about 25% has something much different.
I'll try to grab some of this too. If you create something new, or put it in a second filter it's ok. I don't want to mess up with wrong % of this different captcha.
Comments
pranshua1 send some more sample with correct answer in file name and I give it a try.
here 155 recognized manually samples http://www.mediafire.com/file/38riv312vkd3m8j/cpcnew.zip
Bruteforce no result(
Brazilian NFP deployed new captcha engine generator for this captcha. I guess they have 2 or 3 generators for the same link.
Could you please help me to make the filters/masks work again ?
URL: https://www.nfp.fazenda.sp.gov.br/imagemDinamica.dcontent?0eb3e4716abf43d3aa33f1c56c937d7b
original discussion bellow.
thx
fkomatsu
i have some samples with right names.
ty
If you load just the hard once in that didn't get solved, you get of course a way lower success rate which is not really reflecting the REAL success rate.
Here goes 130 new captchas with answers for nfp.fazenda.gov.sp.br. Most of them are right, but a few are really difficult even for a human being.
http://www.mediafire.com/file/nan57io91aassc1/new_nfp_2017_-_GSA.rar
Jut a remind.
min. lenght of result: 4
max. lenght of result: 4
charset : 0123456789abcdefghijklmnopqrstuvwxyz
I'll try to grab some of this too. If you create something new, or put it in a second filter it's ok. I don't want to mess up with wrong % of this different captcha.