Path to my images CAPTCHA and break CAPTCHA

CAPTCHA and break CAPTCHA are always Turing test propositions.
Both CAPTCHA and break CAPTCHA are competing,

CAPTCHA:

• CAPTCHA Project: http://www.captcha.net/

• Jcaptcha project: http://jcaptcha.sourceforge.net/

Break CAPTCHA

• Http://www.cs.sfu.ca /~ Mori/research/gimpy/

Http://sam.zoy.org/pwntcha/

• Http://www.brains-n-brawn.com/default.aspx? Vdir = aicaptcha

Http://homepages.cs.ncl.ac.uk/jeff.yan/

It feels like a boxing match.

In his image class, Mr. Zhang assigned a big assignment for verification code recognition. This makes me feel lucky to have access to this. As one of the main members of our group, I took the burden of completing this assignment.

Of course, I don't want to be involved in this endless hard AI war. As stated in recognizing objects in adversarial clutter: breaking a visual CAPTCHA, the significance of verification code recognition is that it is helpful for object recognition:
1. For word Extraction of some images, the background noise of the verification code is more demanding and it is not easy to increase the recognition rate;
2. CAPTCHA verification codes are generated open-source and can be used to generate infinit testset. Such datasets are adequate. Extended the previous experiments on mnist.

Reading recognizing objects in adversarial clutter: breaking a visual CAPTCHA is not easy, especially for beginners in the image processing field like me. The authors G. Mori claim that their recognition rate in ez-gimpy is 92% and that in gimpy is 33%. This is a research achievement (http://www.eecs.berkeley.edu/Research/Projects/CS/vision/shape/) in the EECs of ucberkely ).
I have to start with shape context ~ Hehe, the road to image is long, so I will try again and again :)

At the same time, I found an open-source software pwntcha in the project. The author is Sam hocevar and said he saw so many selling verification codes on the market to crack it.ProgramHe changed his mindCodeAll are published. Although it was a technology three years ago, it makes sense because of his powerful recognition capabilities (many of which are 100% ). You can download the code using SVN on the official website, but the code below cannot be used directly, and the author's readme write is not detailed. Some settings are needed after installation.
Here, I also download pwntcha from this site, which has passed my test.

Linux installation:
1. install some libraries (both./configure & make install)
1.1、install imlib2-version.tar.gz
1.2、install sdl-version.tar.gz
1.38.0 install sdl_image-version.tar.gz
1.4、install opencv-version.tar.gz
2. Compile pwntcha
2.1 generate configure script and makefile (CD trunk/&./Bootstrap)
2.2 compile and install (./configure & make install)

At this time, the executable file pwntcha will be generated in the trunk/src/directory. However, the program cannot identify the verification code. You also need to make the following settings:
Create a share directory in the directory where the executable pwntcha is located (trunk/src/by default.
Move the site folder (xanga, VBulletin, tickets, Ticketmaster, Slashdot, scode, phpBB, PayPal...) under trunk/src/to share. That is, each folder under share has a site folder, which enables pwntcha to open the verification code image files for these sites.

After these two steps, you can use pwntcha. Command:
Usage:./pwntcha [Option]... image...
-M, -- Mode & lt; Mode & gt; force operating mode
-S, -- share & lt; Dir & gt; Specify shared dir
-Q, -- quiet do not print information messages
-H, -- help display this help and exit
-V, -- version output version information and exit
Recognition
Output results (click to see the big picture)

Finally, I listed several important professors in the image field with verification codes: