2
# $URL: http://pypng.googlecode.com/svn/trunk/code/exnumpy.py $
6
# Original code created by Mel Raab, modified by David Jones.
9
Example code integrating RGB PNG files, PyPNG and NumPy
10
(abstracted from Mel Raab's functioning code)
13
# http://www.python.org/doc/2.4.4/lib/module-itertools.html
20
''' If you have a PNG file for an RGB image,
21
and want to create a numpy array of data from it.
23
# Read the file "picture.png" from the current directory. The `Reader`
24
# class can take a filename, a file-like object, or the byte data
25
# directly; this suggests alternatives such as using urllib to read
26
# an image from the internet:
27
# png.Reader(file=urllib.urlopen('http://www.libpng.org/pub/png/PngSuite/basn2c16.png'))
28
pngReader=png.Reader(filename='picture.png')
29
# Tuple unpacking, using multiple assignment, is very useful for the
30
# result of asDirect (and other methods).
32
# http://docs.python.org/tutorial/introduction.html#first-steps-towards-programming
33
row_count, column_count, pngdata, meta = pngReader.asDirect()
34
bitdepth=meta['bitdepth']
35
plane_count=meta['planes']
37
# Make sure we're dealing with RGB files
38
assert plane_count == 3
40
''' Boxed row flat pixel:
41
list([R,G,B, R,G,B, R,G,B],
42
[R,G,B, R,G,B, R,G,B])
43
Array dimensions for this example: (2,9)
45
Create `image_2d` as a two-dimensional NumPy array by stacking a
46
sequence of 1-dimensional arrays (rows).
47
The NumPy array mimics PyPNG's (boxed row flat pixel) representation;
48
it will have dimensions ``(row_count,column_count*plane_count)``.
50
# The use of ``numpy.uint16``, below, is to convert each row to a NumPy
51
# array with data type ``numpy.uint16``. This is a feature of NumPy,
52
# discussed further in
53
# http://docs.scipy.org/doc/numpy/user/basics.types.html .
54
# You can use avoid the explicit conversion with
55
# ``numpy.vstack(pngdata)``, but then NumPy will pick the array's data
56
# type; in practice it seems to pick ``numpy.int32``, which is large enough
57
# to hold any pixel value for any PNG image but uses 4 bytes per value when
58
# 1 or 2 would be enough.
59
# --- extract 001 start
60
image_2d = numpy.vstack(itertools.imap(numpy.uint16, pngdata))
62
# Do not be tempted to use ``numpy.asarray``; when passed an iterator
63
# (`pngdata` is often an iterator) it will attempt to create a size 1
64
# array with the iterator as its only element.
65
# An alternative to the above is to create the target array of the right
66
# shape, then populate it row by row:
68
image_2d = numpy.zeros((row_count,plane_count*column_count),
70
for row_index, one_boxed_row_flat_pixels in enumerate(pngdata):
71
image_2d[row_index,:]=one_boxed_row_flat_pixels
77
''' Reconfigure for easier referencing, similar to
78
Boxed row boxed pixel:
79
list([ (R,G,B), (R,G,B), (R,G,B) ],
80
[ (R,G,B), (R,G,B), (R,G,B) ])
81
Array dimensions for this example: (2,3,3)
83
``image_3d`` will contain the image as a three-dimensional numpy
84
array, having dimensions ``(row_count,column_count,plane_count)``.
86
# --- extract 002 start
87
image_3d = numpy.reshape(image_2d,
88
(row_count,column_count,plane_count))
94
''' Convert NumPy image_3d array to PNG image file.
96
If the data is three-dimensional, as it is above, the best thing
97
to do is reshape it into a two-dimensional array with a shape of
98
``(row_count, column_count*plane_count)``. Because a
99
two-dimensional numpy array is an iterator, it can be passed
100
directly to the ``png.Writer.write`` method.
103
row_count, column_count, plane_count = image_3d.shape
104
assert plane_count==3
106
pngfile = open('picture_out.png', 'wb')
108
# This example assumes that you have 16-bit pixel values in the data
109
# array (that's what the ``bitdepth=16`` argument is for).
110
# If you don't, then the resulting PNG file will likely be
111
# very dark. Hey, it's only an example.
112
pngWriter = png.Writer(column_count, row_count,
116
# As of 2009-04-13 passing a numpy array that has an element type
117
# that is a numpy integer type (for example, the `image_3d` array has an
118
# element type of ``numpy.uint16``) generates a deprecation warning.
119
# This is probably a bug in numpy; it may go away in the future.
120
# The code still works despite the warning.
121
# See http://code.google.com/p/pypng/issues/detail?id=44
122
# --- extract 003 start
123
pngWriter.write(pngfile,
124
numpy.reshape(image_3d, (-1, column_count*plane_count)))
125
# --- extract 003 end