dsift.py

import numpy
import os
import tempfile
import shutil
from PIL import Image
import scipy.io as sio

def process_image_dsift(imname, resultname, size=20, steps=10,
                        force_orientation=False, resize=None):
  """Process an image with densely sampled sift descriptors and save the
  results in a file. Optional input: size of features, steps between locations,
  forcing computation of descriptor orientation (False means all are oriented
  upward), tuple for resizing the image."""
  im = Image.open(imname).convert('L')
  if resize is not None:
    im = im.resize(resize)
  m, n = im.size

  temp_dir = tempfile.mkdtemp()
  try:
    if not imname.endswith('pgm'):
      tmpname = os.path.join(temp_dir, 'out_tmp.pgm')
      im.save(tmpname)
      imname = tmpname

    scale = size / 3.0
    x, y = numpy.meshgrid(range(steps, m, steps), range(steps, n, steps))
    xx, yy = x.flatten(), y.flatten()
    tmpframe = os.path.join(temp_dir, 'out_tmp.frame')
    frame = numpy.array([xx, yy, scale * numpy.ones(xx.shape[0]),
                                 numpy.zeros(xx.shape[0])])
    numpy.savetxt(tmpframe, frame.T, fmt='%03.3f')

    cmd = ['sift', imname, '--output=' + resultname,
           '--read-frames=' + tmpframe]
    if force_orientation:
      cmd += ['--orientations']
    os.system(' '.join(cmd))

    # Re-write as .mat file, which loads faster.
    f = numpy.loadtxt(resultname)
    sio.savemat(resultname + '.mat', {'f':f}, oned_as='row')
  finally:
    shutil.rmtree(temp_dir)