• Author: Dov Grobgeld [email protected]
• Created: 2013-12-14 Sat
• Keywords: Python, Halcon
• URL: http://github.com/dov/hirsch
• Version: 0.1

Hirsch is a python binding to the MVTec proprietory image processing library HALCON, see: http://www.halcon.com/ . It has currently been tested with HALCON version 9.0 .

Note about the name. Since HALCON is a proprietary name belonging to MVTec, I have decided to name this binding Hirsch.

The use of Hirsch requires a licence to the HALCON library. Please refer to the HALCON End User Legal Agreement (EULA) for details.

The licence of the Python binding is LGPL3 and of the automatic python generator, GPL3.

See license.md for license details about Hirsch.

There are two steps concerning building the Hirsch binding:

This step parses the Halcon include files and generates python binding. It is run by doing:

python h-parse.py

Note that you will need to edit h-parse.py to set the halcon include directory at the top of the files.

This distribution contains pregenerated bindings for halcon version 9, so this step may be skipped.

Please edit setup.py to specify include and library paths for halcon. Install as standard python modules:

python setup.py install 

Here is an example of how to use Hirsch:

from hirsch import HImage, HDataCode2D

Filename = 'qr.tiff'
img = HImage.ReadImage(Filename)
print "Image size=", img.Width(), img.Height()

QrDetector = HDataCode2D("QR Code","default_parameters","enhanced_recognition")
QRContours,Handles,DataStrings = QrDetector.FindDataCode2d(img,"stop_after_result_num",1)
print DataStrings

A few more examples are available in examples.

No effort has been made yet to do bindings for the built in visualization. Instead the library hirsch.giv provides a simple interface for using the image viewer giv for visualization, see: http://giv.sourceforge.net/giv/index.html .

Here is an example of how to view blobs with blob info with giv.

from hirsch import HImage, HDataCode2D
import hirsch.giv as giv

Filename = 'qrfoo.tif'
img = HImage.ReadImage(Filename)
blobs = img.Threshold(0,128).Connection().ErosionCircle(3)

blobs = [b for b in blobs if b.Circularity() > 0.5]

giv.ViewRegions(blobs,image=img, props=['Circularity','Area'])

Here is the resulting giv image:

When hoovering over the red blobs, the Circularity and the Area properties for those blobs are shown in a popup window.

HImage transparently and quickly converts to and from a numpy.array. I.e. the following code can be used for converting from a 2D numpy.array to an HImage and the other way around:

  my_array = numpy.array(my_himg, copy=0)
  my_himg1 = hirsch.HImage(my_array)

Note that an HImage can be used directly whenever an numpy.array is expected. E.g. to show an HImage with pylab:

  imshow(my_himg)

The option copy=0 to numpy.array() specifies that my_array will be pointing to the same memory as is used by the my_himg. Omitting this option will copy this memory. See numpy.array.

One of the nice aspects of the python binding is that it works transparently with ipython (interactive python). In particular the ipython autocompletion and help system are great tools for working interactively with hirsch. Here is an example of an interactive session:

In [1]: from hirsch import *

In [2]: from hirsch.giv import *

In [3]: img = HImage.Read<tab>
HImage.ReadGraySe    HImage.ReadImage     HImage.ReadSequence

In [3]: img = HImage.ReadImage('test_data/<tab>
test_data/maja.png
test_data/qrfoo.tif

In [3]: img = HImage.ReadImage('test_data/maja.png')

In [5]: img.Ed<tab>
img.EdgesColor        img.EdgesImage        
img.EdgesColorSubPix  img.EdgesSubPix       

# The help text is the comment above the corresponding 
# method in the C++ h-file.
In [5]: img.EdgesSubPix?
Type:       builtin_function_or_method
String Form:<built-in method EdgesSubPix of Halcon.Image object at 0x9c5bf20>
Docstring:
EdgesSubPix(Filter,Alpha,Low,High)

Extract sub-pixel precise edges using Deriche, Lanser, Shen, or Canny
filters.

In [6]: edges = img.EdgesSubPix('canny',2,1,10)

In [7]: len(edges)
Out[7]: 432

In [8]: ViewContours(edges,image=img,color='red',linewidth=2)

In [9]: edges[0]

Out[9]: <Halcon.hxldcont at 0xb76cbde0>

# Note: Need to assign to a variable to use completion of an array.
In [10]: e = edges[0]

In [11]: e.MomentsXld? 
Type:       builtin_function_or_method
String Form:<built-in method MomentsXld of Halcon.hxldcont object at 0xb76cbc80>
Docstring:
MomentsXld()

Geometric moments M20, M02, and M11 of contours or polygons.

In [11]: e.MomentsXld()
Out[11]: (-5002.824723977863, 2913.0275057177496, 10901.533649534855)

In [12]: e.LengthXld()
Out[12]: 56.4611522953062

In [13]: max([e.LengthXld() for e in edges])
Out[13]: 327.5589860358794

In [14]: LongEdges = [e for e in edges if e.LengthXld()>300]

In [15]: ViewContours(LongEdges,image=img,color='red',linewidth=2)

The autogenerated codes resides in .i files. There are two such files for each Hirsch class. One containing the "list" code that define the mapping between python method names, the c-function implementation, and its documentation. The second contains the implementation of the c-functions.

I have tried to make the binding as pythonesque as possible. This means that I have tried to use python structures as much as possible. E.g. for functions taking both HTuple and pod (plain old data) as input, the HTuple versions were skipped.

One problem that I have not satisfactory solved are functions that return scalars by HTuple's. There is no indication in the header files that the returned value is a scalar and not an arbitrary length array. In these cases I return a Python Tuple as a return value and the user must dereference it.

I have succesfully taken care of operator overloading.