The CAPTCHA Breaking Library and Scripting Language provides the necessary tools for quickly creating a program capable of reading text out of an image. The actual job of determining which letter is in a given image is done with the help of Neural Networks, Contour Analysis, and Bitmap Vector Subtraction.
CAPTCHAs that are able to be segmented by color (i.e., each letter is a different color) may first be converted to a perceptive color space where distances between colors are mathematically determined based on how the human eyes perceive color, not how colors are different in the RGB color-space. This allows most multicolor CAPTCHAs to be solved quite trivially.
To get started, check out the hello world tutorial, the language syntax, and the solver walkthrough.
Here is a code snippet written in CBL that breaks a CAPTCHA originally from here.
**********************************************************
* Scott Clayton April 14, 2012 *
**********************************************************
* This script is part of the CBL interpreter: *
* http://code.google.com/p/captcha-breaking-library/ *
**********************************************************
* The CAPTCHA that this script breaks came from: *
* http://www.codeproject.com/Articles/5947/CAPTCHA-Image *
**********************************************************
SetMode, all
SetupSegmenter, BLOB, 4, 14, 8
SetupSolver, MNN, "0123456789", 20, 20, 8, 150, 0.95
Load, "mnn.solver.db"
DefinePreconditions
Resize, 400, 100
Subtract, "merge3.bmp"
Invert
Median, 1
MeanShift, 1, 2, 5
Binarize, 150
ColorFillBlobs, 80, 52
RemoveSmallBlobs, 90, 4, 14
HistogramRotate
Binarize, 200
ColorFillBlobs
EndPreconditions
Solve, %IMAGE%
Here's the CBL GUI running the script you see above on a CAPTCHA:
Be sure to install the Notepad++ syntax highlighting plugin on the downloads page to get syntax highlighting for CBL. Installation instructions are included in the Readme.txt file in the download.
Scott
Exported from my old Google Code repository.