def __img_size_tiff(self, fn):
# We can use the VIPS module here for TIFF, since it can handle all the
# ins and outs of the TIFF image format quite nicely.
from vipsCC import VImage
im = VImage.VImage(fn)
size = (im.Xsize(), im.Ysize())
del im
return size
def get_image_info(filepath, tilewidth):
# First, find the largest dimension of the image
image = VImage.VImage(filepath)
width = image.Xsize()
height = image.Ysize()
largest_dim = width if width > height else height
# Now figure out the number of zooms
zoom_levels = math.ceil(math.log((largest_dim + 1) / (tilewidth), 2)) + 1
return (int(zoom_levels), (width, height))
def __img_size_tiff(self, fn):
# We can use the VIPS module here for TIFF, since it can handle all the
# ins and outs of the TIFF image format quite nicely.
# if we're not dealing with TIFF, we don't need to import a non-core library.
# Since jpeg2000 works by reading the header directly, we've made the choice to
# import this with every call. It's not ideal, but it shouldn't be too bad.
# If you are dealing with TIFF files and want to make a slight optimization you
# can move this import statement to the top of this script.
from vipsCC import VImage
im = VImage.VImage(fn)
size = (im.Xsize(), im.Ysize())
del im
return size
def run():
global img_file, res_prefix, s_factors, n_splits, ovlap
# print(img_file)
# print(res_prefix)
# print(s_factors)
# print(n_splits)
# print(ovlap)
r = ET.Element('meta')
t = ET.SubElement(r, 'file')
t.text = img_file
t = ET.SubElement(r, 'parameters')
t1 = ET.SubElement(t, 'prefix')
t1.text = res_prefix
t1 = ET.SubElement(t, 'shrink')
for s in s_factors:
t2 = ET.SubElement(t1, 'factor')
t2.text = str(s)
t1 = ET.SubElement(t, 'split')
for s in n_splits:
t2 = ET.SubElement(t1, 'tile')
t2.text = str(s)
t1 = ET.SubElement(t, 'overlap')
t1.text = str(ovlap)
img = VImage.VImage(img_file)
t1 = ET.SubElement(r, 'original')
t2 = ET.SubElement(t1, 'width')
t2.text = str(img.Xsize())
t2 = ET.SubElement(t1, 'height')
t2.text = str(img.Ysize())
t2 = ET.SubElement(t1, 'channels')
t2.text = str(img.Bands())
t2 = ET.SubElement(t1, 'xres')
t2.text = str(img.Xres())
t2 = ET.SubElement(t1, 'yres')
t2.text = str(img.Yres())
t2 = ET.SubElement(t1, 'scale')
t2.text = '1.0'
t2 = ET.SubElement(t1, 'tile')
t2.text = '(1, 1)'
path = res_prefix + '/' + os.path.basename(img_file)
if os.path.exists(path):
print('Warning: Overwriting old files!')
else:
os.mkdir(path)
print("ROI detection: ")
# Find the ROI:
# img_scaled = img.shrink(100, 100)
os.spawnv(os.P_WAIT, 'div100', ['./div100', img_file, path + '/small.ppm'])
# save downscaled image - not the best way for going to Scikit Image,
# but for large images we go through disk I/O anyway:
# print(" -saving small version of the image")
# img_scaled.write(path+'/small.ppm')
print(" -read into scikit-learn")
img_scaled = imread(path + '/small.ppm')
# compute a minimal area based on the resolution of the image
# -the image is 100x smaller than the original -> resolution is
print(" -computing mask")
xres, yres = img.Xres() / 100, img.Yres() / 100
min_area = 4 * min(xres, yres) # ~4 mm**2
mask, _ = tissue_region_from_rgb(img_scaled, min_area)
# save the mask:
print(" -saving mask")
imsave(path + '/' + 'mask_div100.pbm', mask)
t2 = ET.SubElement(t1, 'mask')
t2.text = 'mask_div100.pbm'
# coordinates of the ROI encompassing the objects, at 0.01 of original image
print(" -detect ROI")
rmin, cmin, rmax, cmax = bounding_box(mask)
mask = mask[rmin:rmax + 1, cmin:cmax + 1]
# get back to original coordinates, with an approximation of 100 pixels...
rmin *= 100
cmin *= 100
rmax = min((rmax + 1) * 100, img.Ysize())
cmax = min((cmax + 1) * 100, img.Xsize())
t2 = ET.SubElement(t1, 'roi', {
'xmin': str(cmin),
'ymin': str(rmin),
'xmax': str(cmax),
'ymax': str(rmax)
})
print("...end ROI detection.")
# Save initial level 0:
print("Crop ROI and save...")
img_cropped = img.extract_area(cmin, rmin, cmax - cmin + 1,
rmax - rmin + 1)
img_cropped.write(path + '/pyramid-level_0.ppm')
new_width, new_height = img_cropped.Xsize(), img_cropped.Ysize()
img_cropped = None
print("...OK")
mask = None
img = None # done with it
gc.collect()
# Generate the pyramid
t1 = ET.SubElement(r, 'pyramid')
t2 = ET.SubElement(t1, 'level', {
'value': '0',
'file': 'pyramid-level_0.ppm'
})
t2 = ET.SubElement(t1, 'level', {
'value': '1',
'file': 'pyramid-level_1.ppm'
})
t2 = ET.SubElement(t1, 'level', {
'value': '2',
'file': 'pyramid-level_2.ppm'
})
t2 = ET.SubElement(t1, 'level', {
'value': '3',
'file': 'pyramid-level_3.ppm'
})
t2 = ET.SubElement(t1, 'level', {
'value': '4',
'file': 'pyramid-level_4.ppm'
})
t2 = ET.SubElement(t1, 'level', {
'value': '5',
'file': 'pyramid-level_5.ppm'
})
# call external tool:
print("Computing pyramid...")
os.spawnv(os.P_WAIT, 'pyr', [
'./pyr', path + '/pyramid-level_0.ppm', path + '/pyramid',
str(n_levels)
])
print("...done")
k = 0
for l in np.arange(n_levels + 1):
f = s_factors[l]
pt = path + '/' + str(s_factors[l])
if not os.path.exists(pt):
os.mkdir(pt)
t1 = ET.SubElement(r, 'version')
t2 = ET.SubElement(t1, 'scale')
t2.text = str(f)
n_horiz, n_vert = n_splits[k]
t2 = ET.SubElement(t1, 'split')
t2.text = str((n_horiz, n_vert))
# img_scaled = img_cropped.shrink(f, f)
# load the corresponding level in the pyramid:
img_scaled = VImage.VImage(path + '/pyramid-level_' + str(l) + '.ppm')
width = img_scaled.Xsize()
height = img_scaled.Ysize()
w = width / n_horiz
h = height / n_vert
ov = ovlap / 100.0 / 2.0 # ovlap is in % and we need half of it
sv = int(n_vert != 1)
sh = int(n_horiz != 1)
print('Processing scale %d and tile %d,%d' % (f, n_horiz, n_vert))
y0 = 0
for i in np.arange(n_vert):
x0 = 0
if i < n_vert - 1:
y1 = int(y0 + h * (1.0 + sv * ov)) - 1
else:
y1 = height - 1
for j in np.arange(n_horiz):
if j < n_horiz - 1:
x1 = int(x0 + w * (1.0 + sh * ov)) - 1
else:
x1 = width - 1
tile_name = 'tile_' + str(i) + '_' + str(j) + '.' + res_format
res_file = pt + '/' + tile_name
print('Save to' + res_file)
t2 = ET.SubElement(
t1, 'tile', {
'name': tile_name,
'x0': str(x0),
'y0': str(y0),
'x1': str(x1),
'y1': str(y1)
})
#print('x0 = %d, y0 = %d, x1 = %d, y1 = %d, sh = %d, ov = %f; image: %d x %d' % (x0, y0, x1, y1, sh, ov, width, height))
# do the actual work...
img_sub = img_scaled.extract_area(x0, y0, x1 - x0 + 1,
y1 - y0 + 1)
img_sub.write(res_file)
x0 = int(x1 + 1 - 2.0 * w * ov)
y0 = int(y1 + 1 - 2.0 * w * ov)
k += 1
raw_txt = ET.tostring(r, 'utf-8')
reparsed = minidom.parseString(raw_txt)
pp_txt = reparsed.toprettyxml(indent=' ')
#print(pp_txt)
meta_file = open(path + '/meta.xml', 'w')
meta_file.write(pp_txt)
return
def main(opts):
directory = opts['outd']
resize_images = opts['resz']
quality = opts['qual']
tilesize = opts['tsze']
compression = opts['comp']
twid = float(tilesize)
# set the compression options. We only need to munge it
# for jpeg -- all the other options we don't need a quality
# declaration. If one is specified, we ignore it.
if compression == "jpeg":
compression = "{0}:{1}".format(compression, quality)
# Create a directory called "processed" within that directory
# If that directory already exists, fail
if os.path.isdir(os.path.join(directory, 'processed')):
print 'There already is a processed directory! Delete it and try again.'
sys.exit(1)
else:
os.mkdir(os.path.join(directory, 'processed'))
# Store the zooms of the files in a list
# Use another list to store filenames (same indices etc)
max_zoom_list = []
filename_list = []
dimensions_list = []
for dirpath, dirnames, filenames in os.walk(directory):
for filename in filenames:
if filename.startswith("."):
continue
max_zoom, dimensions = get_image_info(os.path.join(directory, filename), twid)
print "file: {0} has a maximum zoom of {1} ({2} zoom levels).".format(filename, (max_zoom - 1), max_zoom)
max_zoom_list.append(max_zoom)
filename_list.append(filename)
dimensions_list.append(dimensions)
# Now get the absolute lowest and highest max zoom
lowest_max_zoom = min(max_zoom_list)
# Store the number of images to figure out how many 0s we need
num_images = len(filename_list)
num_zeroes = len(str(num_images))
# Now figure out which files have a zoom larger than that
for i,filename in enumerate(filename_list):
fn,ext = os.path.splitext(filename)
input_file = os.path.join(directory, filename)
new_fn = fn.replace(' ', '_') # Replaces all spaces with _ because spaces can cause problems in the long run
output_file = os.path.join(directory, 'processed', "{0}.tif".format(new_fn))
print "Processing {0}".format(input_file)
vimage = VImage.VImage(input_file)
# If the image needs to be resized
if max_zoom_list[i] > lowest_max_zoom and resize_images:
print '{0} needs to be resized, resizing and converting now'.format(filename)
# Resize this image to the proper size ... prepend resized_
width, height = dimensions_list[i]
new_width, new_height = resize_image(lowest_max_zoom, width, height, twid)
vimage.resize_linear(new_width, new_height).vips2tiff('{0}:{1},tile:{2}x{2},pyramid'.format(output_file, compression, tilesize))
else:
vimage.vips2tiff('{0}:{1},tile:{2}x{2},pyramid'.format(output_file, compression, tilesize))
# Now print out the max_zoom this document has
print "This document has a max zoom of: {0}".format(lowest_max_zoom)
def __process_tiff(self, input_file, output_file):
from vipsCC import VImage
vimage = VImage.VImage(input_file)
vimage.vips2tiff('{0}:{1},tile:256x256,pyramid'.format(
output_file, self.compression))
del vimage
def image_size(fn):
img = VImage.VImage(fn)
size = (img.Xsize(), img.Ysize())
del img
return size
def run():
global img_file, res_prefix, s_factors, n_splits, ovlap
print(img_file)
print(res_prefix)
print(s_factors)
print(n_splits)
print(ovlap)
r = ET.Element('meta')
t = ET.SubElement(r, 'file')
t.text = img_file
t = ET.SubElement(r, 'parameters')
t1 = ET.SubElement(t, 'prefix')
t1.text = res_prefix
t1 = ET.SubElement(t, 'shrink')
for s in s_factors:
t2 = ET.SubElement(t1, 'factor')
t2.text = str(s)
t1 = ET.SubElement(t, 'split')
for s in n_splits:
t2 = ET.SubElement(t1, 'tile')
t2.text = str(s)
t1 = ET.SubElement(t, 'overlap')
t1.text = str(ovlap)
img = VImage.VImage(img_file)
t1 = ET.SubElement(r, 'original')
t2 = ET.SubElement(t1, 'width')
t2.text = str(img.Xsize())
t2 = ET.SubElement(t1, 'height')
t2.text = str(img.Ysize())
t2 = ET.SubElement(t1, 'channels')
t2.text = str(img.Bands())
t2 = ET.SubElement(t1, 'xres')
t2.text = str(img.Xres())
t2 = ET.SubElement(t1, 'yres')
t2.text = str(img.Yres())
t2 = ET.SubElement(t1, 'scale')
t2.text = '1.0'
t2 = ET.SubElement(t1, 'tile')
t2.text = '(1, 1)'
path = res_prefix + '/' + os.path.basename(img_file)
if os.path.exists(path):
print('Warning: Overwriting old files!')
else:
os.mkdir(path)
k = 0
for f in s_factors:
pt = path + '/' + str(f)
if not os.path.exists(pt):
os.mkdir(pt)
t1 = ET.SubElement(r, 'version')
t2 = ET.SubElement(t1, 'scale')
t2.text = str(f)
n_horiz, n_vert = n_splits[k]
t2 = ET.SubElement(t1, 'split')
t2.text = str((n_horiz, n_vert))
img_scaled = img.shrink(f, f)
width = img_scaled.Xsize()
height = img_scaled.Ysize()
w = width / n_horiz
h = height / n_vert
ov = ovlap / 100.0 / 2.0 # ovlap is in % and we need half of it
sv = int(n_vert != 1)
sh = int(n_horiz != 1)
print('Processing scale %d and tile %d,%d' % (f, n_horiz, n_vert))
y0 = 0
for i in np.arange(n_vert):
x0 = 0
if i < n_vert - 1:
y1 = int(y0 + h * (1.0 + sv * ov)) - 1
else:
y1 = height - 1
for j in np.arange(n_horiz):
if j < n_horiz - 1:
x1 = int(x0 + w * (1.0 + sh * ov)) - 1
else:
x1 = width - 1
tile_name = 'tile_' + str(i) + '_' + str(j) + '.' + res_format
res_file = pt + '/' + tile_name
print('Save to' + res_file)
t2 = ET.SubElement(
t1, 'tile', {
'name': tile_name,
'x0': str(x0),
'y0': str(y0),
'x1': str(x1),
'y1': str(y1)
})
#print('x0 = %d, y0 = %d, x1 = %d, y1 = %d, sh = %d, ov = %f; image: %d x %d' % (x0, y0, x1, y1, sh, ov, width, height))
# do the actual work...
img_sub = img_scaled.extract_area(x0, y0, x1 - x0 + 1,
y1 - y0 + 1)
img_sub.write(res_file)
x0 = int(x1 + 1 - 2.0 * w * ov)
y0 = int(y1 + 1 - 2.0 * w * ov)
k += 1
raw_txt = ET.tostring(r, 'utf-8')
reparsed = minidom.parseString(raw_txt)
pp_txt = reparsed.toprettyxml(indent=' ')
#print(pp_txt)
meta_file = open(path + '/meta.xml', 'w')
meta_file.write(pp_txt)