peterzs / zhang-suen-thinning

An OpenCV/C++ implementation of the Zhang-Suen thinning algorithm along with related pre and post processing steps.

This is an adaptation of the Zhang-Suen thinning algorithm and related pre-processing and post-processing steps as presented in Algorithms for Image Processing and Computer Vision by JR Parker.

Prerequisites

For only C++

• OpenCV

For Python support

• OpenCV, built with Python bindings.

• Boost::Python

• Boost.Numpy

  $ git clone https://github.com/yati-sagade/zhang-suen-thinning.git

  $ cd zhang-suen-thinning

Building

The library exposing the function that performs Zhang-Suen thinning can be built in two ways. For use in C++ programs, the static llibrary libzhangsuen.a can be produced by going into the toplevel directory for the project and typing

$ make

This will produce libzhangsuen.a in the current directory, which can then be used to link with C++ programs using the library.

For use with Python, OpenCV must be installed, built with Python support. Then, the Python extension can be built by doing

$ make python

This will produce zhangsuen.so in python/, which is an extension directly importable from Python:

$ cd python
$ python
Python 2.7.3 (default, Sep  9 2012, 17:41:34) 
[GCC 4.7.1] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> import zhangsuen
>>> zhangsuen.thin
<Boost.Python.function object at 0x12f67e0>
>>>

Reference

C++

void thin(cv::Mat& img,
          bool need_boundary_smoothing=false,
          bool need_acute_angle_emphasis=false,
          bool destair=false)

Python

zhangsuen.thin(img,
               need_boundary_smoothing,
               need_acute_angle_emphasis,
               destair) -> numpy.ndarray

Arguments

img: The input grayscale(ideally binarized) image (A `numpy.ndarray` in
     Python). Thinning is performed in place in C++ and a new `numpy.ndarray`
     object holding the thinned image is returned in Python. 

need_boundary_smoothing: Whether or not to perform the Stentiford Boundary
                         smoothing pre processing to reduce line fuzz. Line
                         fuzz is the name given to spurious projections and
                         extra lines on the skeleton boundaries.

need_acute_angle_emphasis: Whether or not to perform acute-angle emphasis
                           to reduce necking. 

                           Sometimes, a narrow point at the intersection of
                           lines is stretched into a small line segment.
                           This is called "necking".
                           This type of artifacts can be reduced by setting
                           all the black points that tend to choke an acute
                           angle joint to white. This operation is known as
                           acute-angle emphasis.

destair: Whether to perform Holt's destairing pre processing operation.

         Sometimes, after thinning, there still are pixels that could be
         removed. Many of such removable pixels form staircases in oblique
         lines. Holt's post processing, called destairing, attempts to
         delete such points which, when removed, do not affect the
         connectivity or shape information of the skeleton.