Viewing file:  images.py (6.29 KB) -rw-r--r--
Select action/file-type:
 (+) |  (+) |  (+) | Code (+) | Session (+) |  (+) | SDB (+) |  (+) |  (+) |  (+) |  (+) |  (+) |
"""Image/texture implementation code
This module provides the Pan-OpenGL operations required to support OpenGL
image handling. Most of this code is simply boilerplate code that sets
OpenGL parameters such that normal Pythonic assumptions about data-ordering
are met to allow easier interaction with other projects (such as PIL or
Numpy).
Generally speaking, there are 3 pieces of information which control how
an image is processed in the system:
format -- this is the pixel format, such as GL_RGB/GL_RED/GL_ABGR_EXT
dims -- tuple of dimensions for the image, (width,height,depth) order
type -- the storage data-type for the image, normally GL_UNSIGNED_BYTE
when working in Python, but all of the standard OpenGL types for
images can be used if you happen to have your data in some exotic
format.
OpenGL.UNSIGNED_BYTE_IMAGES_AS_STRING -- if this global value is set,
then read of unsigned byte images using glReadPixels and
glGetTexImage produces a string instead of the default array format.
Attributes of Note:
COMPONENT_COUNTS -- used to lookup how many units of a
given storage type are required to store a unit in a given format
TYPE_TO_ARRAYTYPE -- maps Image storage types to their array data-type
constants, i.e. maps GL_UNSIGNED_SHORT_4_4_4_4 to GL_UNSIGNED_SHORT
so that we can use the standard array types for manipulating
image arrays.
RANK_PACKINGS -- commands required to set up default array-transfer
operations for an array of the specified rank.
New image formats and types will need to be registered here to be supported,
this means that extension modules which add image types/formats need to alter
the tables described above!
XXX Should be an API to handle that instead of direct modification.
"""
from OpenGL.raw import GL as simple
from OpenGL import arrays
import OpenGL
import ctypes
def SetupPixelRead( format, dims, type):
"""Setup transfer mode for a read into a numpy array return the array
Calls setupDefaultTransferMode, sets rankPacking and then
returns a createTargetArray for the parameters.
"""
setupDefaultTransferMode()
# XXX this is wrong? dims may grow or it may not, depends on whether
# the format can fit in the type or not, but rank is a property of the
# image itself? Don't know, should test.
rankPacking( len(dims)+1 )
return createTargetArray( format, dims, type )
def setupDefaultTransferMode( ):
"""Set pixel transfer mode to assumed internal structure of arrays
Basically OpenGL-ctypes (and PyOpenGL) assume that your image data is in
non-byte-swapped order, with big-endian ordering of bytes (though that
seldom matters in image data). These assumptions are normally correct
when dealing with Python libraries which expose byte-arrays.
"""
simple.glPixelStorei(simple.GL_PACK_SWAP_BYTES, 0)
simple.glPixelStorei(simple.GL_PACK_LSB_FIRST, 0)
def rankPacking( rank ):
"""Set the pixel-transfer modes for a given image "rank" (# of dims)
Uses RANK_PACKINGS table to issue calls to glPixelStorei
"""
for func,which,arg in RANK_PACKINGS[rank]:
try:
func(which,arg)
except Exception, err:
# XXX should be logging a warning!
pass
def createTargetArray( format, dims, type ):
"""Create storage array for given parameters
If storage type requires > 1 unit per format pixel, then dims will be
extended by 1, so in the common case of RGB and GL_UNSIGNED_BYTE you
will wind up with an array of dims + (3,) dimensions. See
COMPONENT_COUNTS for table which controls which formats produce
larger dimensions. The secondary table TIGHT_PACK_FORMATS overrides
this case, so that image formats registered as TIGHT_PACK_FORMATS
only ever return a dims-shaped value. TIGHT_PACK_FORMATS will raise
ValueErrors if they are used with a format that does not have the same
number of components as they define.
Note that the base storage type must provide a zeros method. The zeros
method relies on their being a registered default array-implementation for
the storage type. The default installation of OpenGL-ctypes will use
Numpy arrays for returning the result.
"""
# calculate the number of storage elements required to store
# a single pixel of format, that's the dimension of the resulting array
componentCount = formatToComponentCount( format )
if componentCount > 1:
if type not in TIGHT_PACK_FORMATS:
# requires multiple elements to store a single pixel (common)
# e.g. byte array (typeBits = 8) with RGB (24) or RGBA (32)
dims += (componentCount, )
elif TIGHT_PACK_FORMATS[ type ] < componentCount:
raise ValueError(
"""Image type: %s supports %s components, but format %s requires %s components"""%(
type,
TIGHT_PACK_FORMATS[ type ],
format,
componentCount,
)
)
arrayType = arrays.GL_CONSTANT_TO_ARRAY_TYPE[ TYPE_TO_ARRAYTYPE.get(type,type) ]
return arrayType.zeros( dims )
def formatToComponentCount( format ):
"""Given an OpenGL image format specification, get components/pixel"""
size = COMPONENT_COUNTS.get( format )
if size is None:
raise ValueError( """Unrecognised image format: %r"""%(format,))
return size
def returnFormat( data, type ):
"""Perform compatibility conversion for PyOpenGL 2.x image-as string results
Uses OpenGL.UNSIGNED_BYTE_IMAGES_AS_STRING to control whether to perform the
conversions.
"""
if OpenGL.UNSIGNED_BYTE_IMAGES_AS_STRING:
if type == simple.GL_UNSIGNED_BYTE:
if hasattr( data, 'tostring' ):
return data.tostring()
elif hasattr( data, 'raw' ):
return data.raw
return data
COMPONENT_COUNTS = {
# Image-format-constant: number-of-components (integer)
}
TYPE_TO_BITS = {
# GL-image-storage-type-constant: number-of-bits (integer)
}
TYPE_TO_ARRAYTYPE = {
# GL-image-storage-type-constant: GL-datatype (constant)
}
TIGHT_PACK_FORMATS = {
}
RANK_PACKINGS = {
# rank (integer): list of (function,**arg) to setup for that rank
}
|