def create_grid(self, path, paint=True):
''' create a grid by detecting the tabular borders
:param path: where to find the image
:param paint: if one should paint a test picture
:return:
'''
# path = blackwhitify(path)
imgfile = path
# create an image processing object with the scanned page
exists = os.path.isfile(path)
if not exists:
logging.info("%s not found, passing" % path)
return None
try:
image_to_process = imgproc.ImageProc(imgfile)
except OSError:
logging.info("%s is damaged" % path)
return None
# detect the lines
logging.info("detecting lines in image file '%s'..." % (imgfile))
with timeit_context('line detecting'):
with timeit_context('hlines'):
lines_hough = image_to_process.detect_lines(canny_low_thresh=900, canny_high_thresh=1030,
canny_kernel_size=3,
hough_rho_res=0.2,
hough_theta_res=np.pi / 20,
hough_votes_thresh=round(0.4 * image_to_process.img_w))
logging.info("found %d lines at all" % len(lines_hough))
with timeit_context('hcluster'):
vertical_clusters = image_to_process.find_clusters(imgproc.DIRECTION_VERTICAL,
find_clusters_1d_break_dist,
dist_thresh=self.MIN_COL_WIDTH / 2)
logging.info("thereof %d vertical clusters" % len(vertical_clusters))
horizontal_clusters = image_to_process.find_clusters(imgproc.DIRECTION_HORIZONTAL,
find_clusters_1d_break_dist,
dist_thresh=self.MIN_ROW_WIDTH / 2)
logging.info("thereof %d horizontal clusters" % len(horizontal_clusters))
vertical_lines = [x[1][0] for x in vertical_clusters]
horizontal_lines = [x[1][0] for x in horizontal_clusters]
grid = make_grid_from_positions(vertical_lines, horizontal_lines) # line_positions[p_num])
n_rows = len(grid)
n_cols = len(grid[0])
logging.info("grid with %d rows, %d columns" % (n_rows, n_cols))
return grid
def extract_data_frame(
page_col_pos,
page_row_pos,
p_num,
img_file,
output_path,
page,
):
grid = make_grid_from_positions(page_col_pos, page_row_pos)
n_rows = len(grid)
n_cols = len(grid[0]) if n_rows > 0 else 0
print(f"> page {p_num}: grid with {n_rows} rows, {n_cols} columns")
output_files_basename = img_file.name.split('.')[0]
page_grids_file = output_path + (output_files_basename + '.grids.json')
print(f"saving page grids JSON file to '{page_grids_file}'")
save_page_grids({p_num: grid}, page_grids_file)
table = fit_texts_into_grid(page['texts'], grid)
return datatable_to_dataframe(table)
if len(
col_texts
) >= n_rows: # there should be at least one text box per row
filtered_col_positions.append(prev_col_x)
last_col_x = col_x
prev_col_x = col_x
# manually add border for the last column because it has very few or no text boxes
filtered_col_positions.append(
filtered_col_positions[-1] +
(rightmost_pos - filtered_col_positions[-1]) / 2)
filtered_col_positions.append(rightmost_pos)
# create the grid
if filtered_col_positions:
grid = make_grid_from_positions(filtered_col_positions, row_positions)
n_rows = len(grid)
n_cols = len(grid[0])
print("> page %d: grid with %d rows, %d columns" %
(p_num, n_rows, n_cols))
page_grids[p_num] = grid
else: # this happens for the first page as there's no table on that
print("> page %d: no table found" % p_num)
# save the page grids
# After you created the page grids, you should then check that they're correct using pdf2xml-viewer's
# loadGridFile() function
prev_col_x = col_positions[0]
for col_x in col_positions[1:]:
col_texts = [t for t in texts_in_table if prev_col_x < t['left'] + t['width']/2 <= col_x]
if len(col_texts) >= n_rows: # there should be at least one text box per row
filtered_col_positions.append(prev_col_x)
last_col_x = col_x
prev_col_x = col_x
# manually add border for the last column because it has very few or no text boxes
filtered_col_positions.append(filtered_col_positions[-1] + (rightmost_pos - filtered_col_positions[-1]) / 2)
filtered_col_positions.append(rightmost_pos)
# create the grid
if filtered_col_positions:
grid = make_grid_from_positions(filtered_col_positions, row_positions)
n_rows = len(grid)
n_cols = len(grid[0])
print("> page %d: grid with %d rows, %d columns" % (p_num, n_rows, n_cols))
page_grids[p_num] = grid
else: # this happens for the first page as there's no table on that
print("> page %d: no table found" % p_num)
# save the page grids
# After you created the page grids, you should then check that they're correct using pdf2xml-viewer's
# loadGridFile() function
page_grids_file = os.path.join(OUTPUTPATH, output_files_basename + '.pagegrids.json')
# image space" -> we scale them
# to "text box space"
# %% Create page grids
# After you created the page grids, you should then check that they're correct using pdf2xml-viewer's
# loadGridFile() function
print("creating page grids for all pages...")
page_grids = {}
for p_num, p in pages.items():
# grid = make_grid_from_positions(col_positions[p_num], row_positions[p_num][-13:])
# img_grid=iproc_obj.draw_grid(col_positions_noscale[p_num],row_positions_noscale[p_num][-13:])
grid = make_grid_from_positions(col_positions[p_num], row_positions[p_num])
img_grid = iproc_obj.draw_grid(col_positions_noscale[p_num],
row_positions_noscale[p_num])
cv2.imwrite(os.path.join(OUTPUTPATH, imgfilebasename + '.grids.png'),
img_grid)
n_rows = len(grid)
n_cols = len(grid[0])
print("> page %d: grid with %d rows, %d columns" % (p_num, n_rows, n_cols))
page_grids[p_num] = grid
page_grids_file = os.path.join(OUTPUTPATH, imgfilebasename + '.pagegrids.json')
print("saving page grids JSON file to '%s'" % page_grids_file)
save_page_grids(page_grids, page_grids_file)
# %% Create data frames (requires pandas library)
has_wide_footer_text = any(t['width'] >= min_footer_text_width
for t in line_texts)
# check if there's at least one wide text or if all of the required words for a footer match
if has_wide_footer_text or all_a_in_b(words_in_footer, line_str):
bottom_y = line_top
break
else:
bottom_y = p['height']
page_rowpos = [y for y in pos_y if top_y <= y <= bottom_y]
print("> page %d: %d lines between [%f, %f]" %
(p_num, len(page_rowpos), top_y, bottom_y))
from pdftabextract.extract import make_grid_from_positions
grid = make_grid_from_positions(page_colpos, page_rowpos)
n_rows = 17
n_cols = 12
print("> page %d: grid with %d rows, %d columns" % (p_num, n_rows, n_cols))
from pdftabextract.common import save_page_grids
page_grids_file = os.path.join(
OUTPUTPATH, output_files_basename + '.pagegrids_p3_only.json')
print("saving page grids JSON file to '%s'" % page_grids_file)
save_page_grids({p_num: grid}, page_grids_file)
from pdftabextract.extract import fit_texts_into_grid, datatable_to_dataframe
datatable = fit_texts_into_grid(p['texts'], grid)
line_heights = np.diff(pos_y)
print(
"> page %d: %d lines between [%f, %f], median text height = %f, median line height = %f, min line height = %f, max line height = %f"
% (p_num, len(pos_y), top_y, bottom_y, median_text_height,
np.median(line_heights), min(line_heights), max(line_heights)))
# %% Create page grids
# After you created the page grids, you should then check that they're correct using pdf2xml-viewer's
# loadGridFile() function
print("creating page grids for all pages...")
page_grids = {}
for p_num, p in pages.items():
grid = make_grid_from_positions(col_positions[p_num],
line_positions[p_num])
img_grid = iproc_obj.draw_grid(
col_positions_noscale[p_num],
map(lambda x: x * page_scaling_y, line_positions[p_num]))
cv2.imwrite(os.path.join(OUTPUTPATH, imgfilebasename + '.grids.png'),
img_grid)
n_rows = len(grid)
n_cols = len(grid[0])
print("> page %d: grid with %d rows, %d columns" % (p_num, n_rows, n_cols))
page_grids[p_num] = grid
page_grids_file = os.path.join(OUTPUTPATH, imgfilebasename + '.pagegrids.json')
print("saving page grids JSON file to '%s'" % page_grids_file)
save_page_grids(page_grids, page_grids_file)
# %% Create data frames (requires pandas library)
else: # this happens when the number of columns was not correctly detected
diffsum = None
if diffsum is None or diffsum > CORRECT_COLS_MIN_DIFFSUM: # correct the columns for this page
print('> page %d: corrected (diffsum was %f)' % (p_num, diffsum))
x_offset = page_col_positions[p_num][0]
corrected_pos = list(col_medians + x_offset)
page_col_positions[p_num] = corrected_pos
#%% Create the page grids from the row and column positions and save them
page_grids = {}
for p_num, col_positions in page_col_positions.items():
# create the grid
row_positions = page_row_positions[p_num]
grid = make_grid_from_positions(col_positions, row_positions)
n_rows = len(grid)
n_cols = len(grid[0])
print("> page %d: grid with %d rows, %d columns" % (p_num, n_rows, n_cols))
page_grids[p_num] = grid
# save the page grids
# After you created the page grids, you should then check that they're correct using pdf2xml-viewer's
# loadGridFile() function
page_grids_file = os.path.join(OUTPUTPATH,
output_files_basename + '.pagegrids.json')
print("saving page grids JSON file to '%s'" % page_grids_file)
def do_tablextract(self, g, pdf_path, p_num): # g is globals
print('Starting tablextract')
camelot_method = 'lattice' #stream/lattice
if self.pdf_type == 'normal':
print(pdf_path, p_num)
if 'tabula' in g.text_pdf_method:
tables = read_pdf(
pdf_path,
pages=[p_num],
multiple_tables=True,
java_options=
'-Dsun.java2d.cmm=sun.java2d.cmm.kcms.KcmsServiceProvider')
for i in range(len(tables)):
table_file_path = '%s/%s-%s' % (self.tables_folder_tabula,
p_num, i)
# tables[i].fillna('').to_html('%s.html' % (table_file_path))
try:
tables[i].fillna('').to_csv('%s.csv' % (table_file_path),
encoding='utf-8')
except:
tables[i].fillna('').to_csv('%s.csv' % (table_file_path),
encoding='cp1252')
if 'camelot' in g.text_pdf_method:
tables = camelot.read_pdf(pdf_path,
flavor=camelot_method,
pages=str(p_num))
for i in range(len(tables)):
# print(tables[0].parsing_report)
table_file_path = '%s/%s-%s.csv' % (self.tables_folder_camelot,
p_num, i)
tables.export(table_file_path, f='csv', compress=False)
else:
if self.doc_type == 'image':
# trying camelot
print('Doing camelot-stream')
camelot_method = 'stream' #stream/lattice
tables = camelot.read_pdf(pdf_path,
flavor=camelot_method,
pages=str(p_num))
for i in range(len(tables)):
# print(tables[0].parsing_report)
table_file_path = '%s/%s-%s.csv' % (self.tables_folder_camelot,
p_num, i)
tables.export(table_file_path, f='csv', compress=False)
# Trying pdftabextract
filename = os.path.basename(pdf_path).split('.')[0].split('/')[0]
DATAPATH = self.images_folder # 'data/'
INPUT_XML = '%s/%s.xml' % (self.images_folder, filename)
os.system("pdftohtml -c -hidden -xml -enc UTF-8 -f %s -l %s %s %s" %
(p_num, p_num, pdf_path, INPUT_XML))
# os.system("pdftohtml -c -hidden -f %s -l %s %s %s/%s.html" % (p_num, p_num, pdf_path, self.html_folder, filename))
# Load the XML that was generated with pdftohtml
xmltree, xmlroot = read_xml(INPUT_XML)
# parse it and generate a dict of pages
pages = parse_pages(xmlroot)
# print(pages[p_num]['texts'][0])
p = pages[p_num]
# Detecting lines
if self.doc_type == 'image':
imgfilebasename = '%s-%s_1' % (filename, p_num)
imgfile = self.file_path
elif self.doc_type == 'pdf':
try:
imgfilebasename = '%s-%s_1' % (filename, p_num)
imgfile = '%s/%s-%s_1.png' % (DATAPATH, filename, p_num)
except:
imgfilebasename = filename + str(p_num)
imgfile = '%s/%s-%s_1.png' % (DATAPATH, filename, p_num)
print("\npage %d: detecting lines in image file '%s'..." %
(p_num, imgfile))
# create an image processing object with the scanned page
iproc_obj = imgproc.ImageProc(imgfile)
# calculate the scaling of the image file in relation to the text boxes coordinate system dimensions
page_scaling_x = iproc_obj.img_w / p['width'] # scaling in X-direction
page_scaling_y = iproc_obj.img_h / p[
'height'] # scaling in Y-direction
# detect the lines
lines_hough = iproc_obj.detect_lines(canny_kernel_size=3,
canny_low_thresh=50,
canny_high_thresh=150,
hough_rho_res=1,
hough_theta_res=np.pi / 500,
hough_votes_thresh=round(
0.2 * iproc_obj.img_w))
print("> found %d lines" % len(lines_hough))
# helper function to save an image
def save_image_w_lines(iproc_obj, imgfilebasename):
img_lines = iproc_obj.draw_lines(orig_img_as_background=True)
img_lines_file = os.path.join(
self.temp_folder, '%s-lines-orig.png' % imgfilebasename)
print("> saving image with detected lines to '%s'" %
img_lines_file)
cv2.imwrite(img_lines_file, img_lines)
save_image_w_lines(iproc_obj, imgfilebasename)
# find rotation or skew
# the parameters are:
# 1. the minimum threshold in radians for a rotation to be counted as such
# 2. the maximum threshold for the difference between horizontal and vertical line rotation (to detect skew)
# 3. an optional threshold to filter out "stray" lines whose angle is too far apart from the median angle of
# all other lines that go in the same direction (no effect here)
rot_or_skew_type, rot_or_skew_radians = iproc_obj.find_rotation_or_skew(
radians(0.5), # uses "lines_hough"
radians(1),
omit_on_rot_thresh=radians(0.5))
# rotate back or deskew text boxes
needs_fix = True
if rot_or_skew_type == ROTATION:
print("> rotating back by %f°" % -degrees(rot_or_skew_radians))
rotate_textboxes(p, -rot_or_skew_radians, pt(0, 0))
elif rot_or_skew_type in (SKEW_X, SKEW_Y):
print("> deskewing in direction '%s' by %f°" %
(rot_or_skew_type, -degrees(rot_or_skew_radians)))
deskew_textboxes(p, -rot_or_skew_radians, rot_or_skew_type,
pt(0, 0))
else:
needs_fix = False
print("> no page rotation / skew found")
if needs_fix:
# rotate back or deskew detected lines
lines_hough = iproc_obj.apply_found_rotation_or_skew(
rot_or_skew_type, -rot_or_skew_radians)
save_image_w_lines(iproc_obj, imgfilebasename + '-repaired')
# save repaired XML (i.e. XML with deskewed textbox positions)
repaired_xmlfile = os.path.join(self.temp_folder,
filename + '.repaired.xml')
print("saving repaired XML file to '%s'..." % repaired_xmlfile)
xmltree.write(repaired_xmlfile)
# Clustering vertical lines
# cluster the detected *vertical* lines using find_clusters_1d_break_dist as simple clustering function
# (break on distance MIN_COL_WIDTH/2)
# additionally, remove all cluster sections that are considered empty
# a cluster is considered empty when the number of text boxes in it is below 10% of the median number of text boxes
# per cluster section
MIN_COL_WIDTH = g.MIN_COL_WIDTH # minimum width of a column in pixels, measured in the scanned pages
vertical_clusters = iproc_obj.find_clusters(
imgproc.DIRECTION_VERTICAL,
find_clusters_1d_break_dist,
remove_empty_cluster_sections_use_texts=p[
'texts'], # use this page's textboxes
remove_empty_cluster_sections_n_texts_ratio=0.1, # 10% rule
remove_empty_cluster_sections_scaling=
page_scaling_x, # the positions are in "scanned image space" -> we scale them to "text box space"
dist_thresh=MIN_COL_WIDTH / 2)
print("> found %d clusters" % len(vertical_clusters))
# draw the clusters
img_w_clusters = iproc_obj.draw_line_clusters(
imgproc.DIRECTION_VERTICAL, vertical_clusters)
save_img_file = os.path.join(
self.temp_folder, '%s-vertical-clusters.png' % imgfilebasename)
print("> saving image with detected vertical clusters to '%s'" %
save_img_file)
cv2.imwrite(save_img_file, img_w_clusters)
# Clustering horizontal lines
# cluster the detected *horizontal* lines using find_clusters_1d_break_dist as simple clustering function
# (break on distance MIN_ROW_WIDTH/2)
# additionally, remove all cluster sections that are considered empty
# a cluster is considered empty when the number of text boxes in it is below 10% of the median number of text boxes
# per cluster section
MIN_ROW_WIDTH = g.MIN_ROW_WIDTH # minimum width of a row in pixels, measured in the scanned pages
horizontal_clusters = iproc_obj.find_clusters(
imgproc.DIRECTION_HORIZONTAL,
find_clusters_1d_break_dist,
remove_empty_cluster_sections_use_texts=p[
'texts'], # use this page's textboxes
remove_empty_cluster_sections_n_texts_ratio=0.1, # 10% rule
remove_empty_cluster_sections_scaling=
page_scaling_y, # the positions are in "scanned image space" -> we scale them to "text box space"
dist_thresh=MIN_ROW_WIDTH / 2)
print("> found %d clusters" % len(horizontal_clusters))
# draw the clusters
img_w_clusters_hoz = iproc_obj.draw_line_clusters(
imgproc.DIRECTION_HORIZONTAL, horizontal_clusters)
save_img_file = os.path.join(
self.temp_folder, '%s-horizontal-clusters.png' % imgfilebasename)
print("> saving image with detected vertical clusters to '%s'" %
save_img_file)
cv2.imwrite(save_img_file, img_w_clusters_hoz)
page_colpos = np.array(
calc_cluster_centers_1d(vertical_clusters)) / page_scaling_x
print('found %d column borders:' % len(page_colpos))
print(page_colpos)
page_rowpos = np.array(
calc_cluster_centers_1d(horizontal_clusters)) / page_scaling_y
print('found %d row borders:' % len(page_rowpos))
print(page_rowpos)
# right border of the second column
col2_rightborder = page_colpos[2]
# calculate median text box height
median_text_height = np.median([t['height'] for t in p['texts']])
# get all texts in the first two columns with a "usual" textbox height
# we will only use these text boxes in order to determine the line positions because they are more "stable"
# otherwise, especially the right side of the column header can lead to problems detecting the first table row
text_height_deviation_thresh = median_text_height / 2
texts_cols_1_2 = [
t for t in p['texts'] if t['right'] <= col2_rightborder
and abs(t['height'] -
median_text_height) <= text_height_deviation_thresh
]
# get all textboxes' top and bottom border positions
borders_y = border_positions_from_texts(texts_cols_1_2,
DIRECTION_VERTICAL)
# break into clusters using half of the median text height as break distance
clusters_y = find_clusters_1d_break_dist(
borders_y, dist_thresh=median_text_height / 2)
clusters_w_vals = zip_clusters_and_values(clusters_y, borders_y)
# for each cluster, calculate the median as center
pos_y = calc_cluster_centers_1d(clusters_w_vals)
pos_y.append(p['height'])
print('number of line positions:', len(pos_y))
pttrn_table_row_beginning = re.compile(
r'^[\d Oo][\d Oo]{2,} +[A-ZÄÖÜ]')
# 1. try to find the top row of the table
texts_cols_1_2_per_line = split_texts_by_positions(
texts_cols_1_2,
pos_y,
DIRECTION_VERTICAL,
alignment='middle',
enrich_with_positions=True)
# go through the texts line per line
for line_texts, (line_top, line_bottom) in texts_cols_1_2_per_line:
line_str = join_texts(line_texts)
if pttrn_table_row_beginning.match(
line_str
): # check if the line content matches the given pattern
top_y = line_top
break
else:
top_y = 0
print('Top_y: %s' % top_y)
# hints for a footer text box
words_in_footer = ('anzeige', 'annahme', 'ala')
# 2. try to find the bottom row of the table
min_footer_text_height = median_text_height * 1.5
min_footer_y_pos = p['height'] * 0.7
# get all texts in the lower 30% of the page that have are at least 50% bigger than the median textbox height
bottom_texts = [
t for t in p['texts'] if t['top'] >= min_footer_y_pos
and t['height'] >= min_footer_text_height
]
bottom_texts_per_line = split_texts_by_positions(
bottom_texts,
pos_y + [p['height']], # always down to the end of the page
DIRECTION_VERTICAL,
alignment='middle',
enrich_with_positions=True)
# go through the texts at the bottom line per line
page_span = page_colpos[-1] - page_colpos[0]
min_footer_text_width = page_span * 0.8
for line_texts, (line_top, line_bottom) in bottom_texts_per_line:
line_str = join_texts(line_texts)
has_wide_footer_text = any(t['width'] >= min_footer_text_width
for t in line_texts)
# check if there's at least one wide text or if all of the required words for a footer match
if has_wide_footer_text or all_a_in_b(words_in_footer, line_str):
bottom_y = line_top
break
else:
bottom_y = p['height']
print(bottom_y)
print(pos_y)
# finally filter the line positions so that only the lines between the table top and bottom are left
print(page_rowpos)
print("> page %d: %d lines between [%f, %f]" %
(p_num, len(page_rowpos), top_y, bottom_y))
def subsequent_pairs(l):
"""
Return subsequent pairs of values in a list <l>, i.e. [(x1, x2), (x2, x3), (x3, x4), .. (xn-1, xn)] for a
list [x1 .. xn]
"""
return [(l[i - 1], v) for i, v in enumerate(l) if i > 0]
# page_rowpos = [y for y in pos_y if top_y <= y <= bottom_y]
print(page_colpos, page_rowpos)
grid = make_grid_from_positions(page_colpos, page_rowpos)
# print(grid)
n_rows = len(grid)
n_cols = len(grid[0])
print("> page %d: grid with %d rows, %d columns" %
(p_num, n_rows, n_cols))
page_grids_file = os.path.join(self.temp_folder,
filename + '_pagegrids.json')
print("saving page grids JSON file to '%s'" % page_grids_file)
save_page_grids({p_num: grid}, page_grids_file)
datatable = fit_texts_into_grid(p['texts'], grid)
df = datatable_to_dataframe(datatable)
# print(df.head(n=2))
csv_output_file = os.path.join(self.tables_folder, filename + '.csv')
print("saving extracted data to '%s'" % csv_output_file)
df.to_csv(csv_output_file, index=False, header=False)
line_heights = np.diff(pos_y)
print("> page %d: %d lines between [%f, %f], median text height = %f, median line height = %f, min line height = %f, max line height = %f"
% (p_num, len(pos_y), top_y, bottom_y,
median_text_height, np.median(line_heights),
min(line_heights), max(line_heights)))
#%% Create page grids
# After you created the page grids, you should then check that they're correct using pdf2xml-viewer's
# loadGridFile() function
print("creating page grids for all pages...")
page_grids = {}
for p_num, p in pages.items():
grid = make_grid_from_positions(col_positions[p_num], line_positions[p_num])
n_rows = len(grid)
n_cols = len(grid[0])
print("> page %d: grid with %d rows, %d columns" % (p_num, n_rows, n_cols))
page_grids[p_num] = grid
page_grids_file = os.path.join(OUTPUTPATH, output_files_basename + '.pagegrids.json')
print("saving page grids JSON file to '%s'" % page_grids_file)
save_page_grids(page_grids, page_grids_file)
#%% Create data frames (requires pandas library)
# For sake of simplicity, we will just fit the text boxes into the grid, merge the texts in their cells and
# output the result.
# Normally, at this step you will need to do some error correction / parsing as some text boxes are not correctly
# detected during OCR (they might span over multiple columns or numbers are incorrectly detected as letters).
diffsum = None
if diffsum is None or diffsum > CORRECT_COLS_MIN_DIFFSUM: # correct the columns for this page
print('> page %d: corrected (diffsum was %f)' % (p_num, diffsum))
x_offset = page_col_positions[p_num][0]
corrected_pos = list(col_medians + x_offset)
page_col_positions[p_num] = corrected_pos
#%% Create the page grids from the row and column positions and save them
page_grids = {}
for p_num, col_positions in page_col_positions.items():
# create the grid
row_positions = page_row_positions[p_num]
grid = make_grid_from_positions(col_positions, row_positions)
n_rows = len(grid)
n_cols = len(grid[0])
print("> page %d: grid with %d rows, %d columns" % (p_num, n_rows, n_cols))
page_grids[p_num] = grid
# save the page grids
# After you created the page grids, you should then check that they're correct using pdf2xml-viewer's
# loadGridFile() function
page_grids_file = os.path.join(OUTPUTPATH, output_files_basename + '.pagegrids.json')
print("saving page grids JSON file to '%s'" % page_grids_file)
save_page_grids(page_grids, page_grids_file)
