How to write PIL image filter for plain pgm format?

The way I currently deal with this is through numpy:

1. Read image into a 2D numpy array. You don’t need to use numpy, but I’ve found it easier to use than the regular Python 2D arrays
2. Convert 2D numpy array into PIL.Image object using PIL.Image.fromarray

If you insist on using PIL.Image.open, you could write a wrapper that attempts to load a PGM file first (by looking at the header). If it’s a PGM, load the image using the steps above, otherwise just hands off responsibility to PIL.Image.open.

Here’s some code that I use to get a PBM image into a numpy array.

import re
import numpy

def pbm2numpy(filename):
    """
    Read a PBM into a numpy array.  Only supports ASCII PBM for now.
    """
    fin = None
    debug = True

    try:
        fin = open(filename, 'r')

        while True:
            header = fin.readline().strip()
            if header.startswith('#'):
                continue
            elif header == 'P1':
                break
            elif header == 'P4':
                assert False, 'Raw PBM reading not implemented yet'
            else:
                #
                # Unexpected header.
                #
                if debug:
                    print 'Bad mode:', header
                return None

        rows, cols = 0, 0
        while True:
            header = fin.readline().strip()
            if header.startswith('#'):
                continue

            match = re.match('^(d+) (d+)$', header)
            if match == None:
                if debug:
                    print 'Bad size:', repr(header)
                return None

            cols, rows = match.groups()
            break

        rows = int(rows)
        cols = int(cols)

        assert (rows, cols) != (0, 0)

        if debug:
            print 'Rows: %d, cols: %d' % (rows, cols)

        #
        # Initialise a 2D numpy array 
        #
        result = numpy.zeros((rows, cols), numpy.int8)

        pxs = []

        # 
        # Read to EOF.
        # 
        while True:
            line = fin.readline().strip()
            if line == '':
                break

            for c in line:
                if c == ' ':
                    continue

                pxs.append(int(c))

        if len(pxs) != rows*cols:
            if debug:
                print 'Insufficient image data:', len(pxs)
            return None

        for r in range(rows):
            for c in range(cols):
                #
                # Index into the numpy array and set the pixel value.
                #
                result[r, c] = pxs[r*cols + c]

        return result

    finally:
        if fin != None:
            fin.close()
        fin = None

    return None

You will have to modify it slightly to fit your purposes, namely:

• Deal with P2 (ASCII, greyscale) instead of P1 (ASCII, bilevel).
• Use a different container if you’re not using numpy. Normal Python 2D arrays will work just fine.

EDIT

Here is how I would handle a wrapper:

def pgm2pil(fname):
    #
    # This method returns a PIL.Image.  Use pbm2numpy function above as a
    # guide.  If it can't load the image, it returns None.
    #
    pass
    
def wrapper(fname):
    pgm = pgm2pil(fname)

    if pgm is not None:
        return pgm
    return PIL.Image.open(fname)

#
# This is the line that "adds" the wrapper
#
PIL.Image.open = wrapper

I didn’t write pgm2pil because it’s going to be very similar to pgm2numpy. The only difference will be that it’s storing the result in a PIL.Image as opposed to a numpy array. I also didn’t test the wrapper code (sorry, a bit short on time at the moment) but it’s a fairly common approach so I expect it to work.

Now, it sounds like you want other applications that use PIL for image loading to be able to handle PGMs. It’s possible using the above approach, but you need to be sure that the above wrapper code gets added before the first call to PIL.Image.open. You can make sure that happens by adding the wrapper source code to the PIL source code (if you have access).