def get_pt(all_words, height, width, words):
pt = []
for word in all_words:
if word['text'].isdigit():
if 0 < int(word['text']) < 33:
cad = {
'text': word['text'],
'x1': word['left'],
'y1': word['top'],
'x2': word['left'] + word['width'],
'y2': word['top'] + word['height']
}
neighbours = Neighbour.find_neighbour(cad, words,
int(width * 0.3),
int(height * 0.02),
width, height)
neighbours_text = [
it['text'].lower().replace('.', '') for it in neighbours
]
if 'net' in neighbours_text or 'terms' in neighbours_text or (
'due' in neighbours_text
and 'days' in neighbours_text):
pt.append({
'text': word['text'],
'x1': word['left'],
'y1': word['top'],
'x2': word['left'] + word['width'],
'y2': word['top'] + word['height']
})
return pt
def __init__(self, split_name='train'):
""" Initialize the dataset with preprocessing """
cached_data_path = config.OUTPUT_DIR / f"cached_data_{split_name}.pickle"
if cached_data_path.exists():
print("Preprocessed data available, Loading data from cache...")
with open(cached_data_path, "rb") as f:
cached_data = pickle.load(f)
classes_count = cached_data['count']
class_mapping = cached_data['mapping']
print("\nClass Mapping:", class_mapping)
print("Classs counts:", classes_count)
_data = cached_data['data']
self.vocab = cached_data['vocab']
self.field_ids, self.candidate_cords, self.neighbours, self.neighbour_cords, self.mask, self.labels = _data
else:
print("Preprocessed data not available")
annotation, classes_count, class_mapping = xml_parser.get_data(config.XML_DIR, split_name)
print("Class Mapping:", class_mapping)
print("Classs counts:", classes_count)
annotation = candidate.attach_candidate(annotation, config.CANDIDATE_DIR)
annotation, self.vocab = Neighbour.attach_neighbour(annotation, config.OCR_DIR, config.IMAGE_DIR, vocab_size=config.VOCAB_SIZE)
annotation = op.normalize_positions(annotation)
_data = preprocess.parse_input(annotation, class_mapping, config.NEIGHBOURS, self.vocab)
self.field_ids, self.candidate_cords, self.neighbours, self.neighbour_cords, self.mask, self.labels = _data
cached_data = {'count': classes_count, "mapping": class_mapping, 'vocab': self.vocab, 'data': _data}
print("Saving Cache..")
with open(cached_data_path, 'wb') as f:
pickle.dump(cached_data, f, protocol=pickle.HIGHEST_PROTOCOL)
print("Done !!")
def get_subtotal(all_words, height, width, words):
amounts = []
for word in all_words:
if word['text'][0] == '0' and word['text'].isdigit():
continue
try:
formatted_word = re.sub(r'[$,]', '', word['text'])
if not any(char.isdigit() for char in formatted_word):
continue
if '-' in word['text']:
continue
if re.search('[a-zA-Z]', word['text']):
continue
if '/' in word['text']:
continue
if len(word['text']) > 15:
continue
if word['text'].isdigit() and len(word['text']) > 4:
continue
if '(' in word['text']:
continue
if '%' in word['text']:
continue
if len(word['text'].split('.')) == 3:
if len(word['text'].split('.')[2]) == 4:
continue
cad = {
'text': word['text'],
'x1': word['left'],
'y1': word['top'],
'x2': word['left'] + word['width'],
'y2': word['top'] + word['height']
}
neighbours = Neighbour.find_neighbour(cad, words, int(width * 0.5),
int(height * 0.01), width,
height)
neighbours_text = [it['text'].lower() for it in neighbours]
if 'subtotal' in neighbours_text:
amounts.append({
'text': word['text'],
'x1': word['left'],
'y1': word['top'],
'x2': word['left'] + word['width'],
'y2': word['top'] + word['height']
})
except ValueError:
continue
return amounts
def get_invoice_nums(all_words, height, width, words):
inv_nums = []
invoice_no_re = r'^[0-9a-zA-Z-]+$'
for word in all_words:
if not re.search('\d', word['text']):
continue
if len(word['text']) < 2:
continue
if len(word['text'].split('-')) > 2:
continue
result = re.findall(invoice_no_re, word['text'])
if result:
cad = {
'text': word['text'],
'x1': word['left'],
'y1': word['top'],
'x2': word['left'] + word['width'],
'y2': word['top'] + word['height']
}
neighbours = Neighbour.find_neighbour(cad, words, int(width * 0.3),
int(height * 0.02), width,
height)
neighbours_text_clean = [
it['text'].lower().replace('#', '').replace(':', '')
for it in neighbours
]
neighbours_text = [it['text'].lower() for it in neighbours]
if not neighbours:
inv_nums.append({
'text': word['text'],
'x1': word['left'],
'y1': word['top'],
'x2': word['left'] + word['width'],
'y2': word['top'] + word['height']
})
if 'invoice' in neighbours_text_clean or '#' in neighbours_text:
inv_nums.append({
'text': word['text'],
'x1': word['left'],
'y1': word['top'],
'x2': word['left'] + word['width'],
'y2': word['top'] + word['height']
})
return inv_nums
def attach_neighbour_candidates(width, height, ocr_data, candidates):
empty_index = [i for i, ele in enumerate(ocr_data['text']) if ele == ""]
for key in ocr_data.keys():
ocr_data[key] = [
j for i, j in enumerate(ocr_data[key]) if i not in empty_index
]
words = []
for txt, x, y, w, h in zip(ocr_data['text'], ocr_data['left'],
ocr_data['top'], ocr_data['width'],
ocr_data['height']):
x2 = x + w
y2 = y + h
words.append({'text': txt, 'x1': x, 'y1': y, 'x2': x2, 'y2': y2})
x_offset = int(width * 0.1)
y_offset = int(height * 0.1)
for cls, both_cads in candidates.items():
for cad in both_cads:
neighbours = Neighbour.find_neighbour(cad, words, x_offset,
y_offset, width, height)
cad['neighbours'] = neighbours
return candidates
def attach_neighbour_candidates(width, height, ocr_data, candidates):
# empty_index = [i for i, ele in enumerate(ocr_data['text']) if ele == ""]
# for key in ocr_data.keys():
# ocr_data[key] = [j for i, j in enumerate(ocr_data[key]) if i not in empty_index]
# words = []
# for txt, x, y, w, h in zip(ocr_data['text'], ocr_data['left'], ocr_data['top'], ocr_data['width'],
# ocr_data['height']):
# x2 = x + w
# y2 = y + h
# words.append({'text': txt, 'x1': x, 'y1': y, 'x2': x2, 'y2': y2})
BlockType_word = [
item for item in ocr_data['Blocks'] if item['BlockType'] == 'WORD'
]
words = []
for item in BlockType_word:
words.append({
'text':
item['Text'],
'x1':
item['Geometry']['BoundingBox']['Left'] * width,
'y1':
item['Geometry']['BoundingBox']['Top'] * height,
'x2':
item['Geometry']['BoundingBox']['Left'] * width +
item['Geometry']['BoundingBox']['Width'] * width,
'y2':
item['Geometry']['BoundingBox']['Height'] * height +
item['Geometry']['BoundingBox']['Top'] * height
})
x_offset = int(width * 0.3)
y_offset = int(height * 0.02)
for cls, both_cads in candidates.items():
for cad in both_cads:
neighbours = Neighbour.find_neighbour(cad, words, x_offset,
y_offset, width, height)
cad['neighbours'] = neighbours
return candidates
def get_ship_to(lines, height, width, words):
ship_to = []
regexp = ".+ [0-9]{1,4} .+ [A-Z]{2} [0-9]{5}"
coord_left = []
for l in lines:
coord_left.append(l['bbox'][0])
dups = defaultdict(list)
for i, e in enumerate(coord_left):
dups[e].append(i)
index_list = []
x1_list = []
y1_list = []
x2_list = []
y2_list = []
key_list = []
for key in sorted(dups):
if not index_list:
key_list.append(key)
index_list.append(dups[key])
index_list = flatten_list(index_list)
for item in index_list:
x1_list.append(lines[item]['bbox'][0])
y1_list.append(lines[item]['bbox'][1])
x2_list.append(lines[item]['bbox'][2])
y2_list.append(lines[item]['bbox'][3])
else:
if key - key_list[-1] == 1:
key_list.append(key)
index_list.append(dups[key])
index_list = flatten_list(index_list)
index_list = sorted(index_list)
pop_ind = []
rem_list = []
for ind in range(1, len(index_list)):
delta_y1 = (lines[index_list[ind]]['bbox'][1] - \
lines[index_list[ind - 1]]['bbox'][1])/height
if delta_y1 < 0.021:
pop_ind.append(index_list[ind - 1])
pop_ind.append(index_list[ind])
pop_ind = list(set(pop_ind))
text_0 = ' '.join([lines[i]['text'] for i in pop_ind])
text_0 = re.sub(' +', ' ', text_0)
x1_list.append(lines[index_list[ind - 1]]['bbox'][0])
y1_list.append(lines[index_list[ind - 1]]['bbox'][1])
x2_list.append(lines[index_list[ind - 1]]['bbox'][2])
y2_list.append(lines[index_list[ind - 1]]['bbox'][3])
bb_coord_0 = [
min(x1_list),
min(y1_list),
max(x2_list),
max(y2_list)
]
if re.findall(regexp, text_0):
cad = {
'text': text_0,
'x1': bb_coord_0[0],
'y1': bb_coord_0[1],
'x2': bb_coord_0[2],
'y2': bb_coord_0[3]
}
neighbours = Neighbour.find_neighbour(
cad, words, int(width * 0.2),
int(height * 0.01), width, height)
neighbours_text = [
it['text'].lower() for it in neighbours
]
if 'ship' in neighbours_text and 'bill' not in neighbours_text:
ship_to.append(cad)
else:
txt = lines[index_list[ind - 1]]['text']
x1 = [lines[index_list[ind - 1]]['bbox'][0]]
y1 = [lines[index_list[ind - 1]]['bbox'][1]]
x2 = [lines[index_list[ind - 1]]['bbox'][2]]
y2 = [lines[index_list[ind - 1]]['bbox'][3]]
bb_coord = [min(x1), min(y1), max(x2), max(y2)]
rem_list.append(index_list[ind - 1])
if re.findall(regexp, txt):
cad = {
'text': txt,
'x1': bb_coord[0],
'y1': bb_coord[1],
'x2': bb_coord[2],
'y2': bb_coord[3]
}
neighbours = Neighbour.find_neighbour(
cad, words, int(width * 0.2),
int(height * 0.01), width, height)
neighbours_text = [
it['text'].lower() for it in neighbours
]
if 'ship' in neighbours_text and 'bill' not in neighbours_text:
ship_to.append(cad)
txt = lines[index_list[-1]]['text']
x1 = [lines[index_list[-1]]['bbox'][0]]
y1 = [lines[index_list[-1]]['bbox'][1]]
x2 = [lines[index_list[-1]]['bbox'][2]]
y2 = [lines[index_list[-1]]['bbox'][3]]
bb_coord = [min(x1), min(y1), max(x2), max(y2)]
if re.findall(regexp, txt):
rem_list.append(index_list[-1])
cad = {
'text': txt,
'x1': bb_coord[0],
'y1': bb_coord[1],
'x2': bb_coord[2],
'y2': bb_coord[3]
}
neighbours = Neighbour.find_neighbour(
cad, words, int(width * 0.2), int(height * 0.01),
width, height)
neighbours_text = [it['text'].lower() for it in neighbours]
if 'ship' in neighbours_text and 'bill' not in neighbours_text:
ship_to.append(cad)
for ind in rem_list:
index_list.remove(ind)
else:
if len(index_list) == 1:
index_list = flatten_list(index_list)
text = ' '.join([lines[i]['text'] for i in index_list])
text = re.sub(' +', ' ', text)
coord = [
min(x1_list),
min(y1_list),
max(x2_list),
max(y2_list)
]
if re.findall(regexp, text):
cad = {
'text': text,
'x1': coord[0],
'y1': coord[1],
'x2': coord[2],
'y2': coord[3]
}
neighbours = Neighbour.find_neighbour(
cad, words, int(width * 0.2), int(height * 0.01),
width, height)
neighbours_text = [it['text'].lower() for it in neighbours]
if 'ship' in neighbours_text and 'bill' not in neighbours_text:
ship_to.append(cad)
key_list = [key]
index_list = [dups[key]]
for item in dups[key]:
x1_list = [lines[item]['bbox'][0]]
y1_list = [lines[item]['bbox'][1]]
x2_list = [lines[item]['bbox'][2]]
y2_list = [lines[item]['bbox'][3]]
return ship_to
def due_dates(all_text, all_words, height, width, words):
dates, all_dates = [], []
indices = []
matches = search_dates(all_text)
for match in matches:
text = match[0]
if len(text.strip()) < 8:
continue
if '$' in text:
continue
if 'day' in text:
continue
if '%' in text:
continue
if '#' in text:
continue
if ':' in text:
continue
if 'hour' in text.lower():
continue
if (len(text) > 10) and not (re.search('[a-zA-Z]', text)):
continue
if len(text) > 15:
continue
if text.isdigit():
continue
if '(' in text:
continue
token_length = len(text.split(' '))
idx = all_text.find(match[0])
text_len = len(text)
index = len(all_text[:idx].strip().split(' '))
if idx == 0:
index = 0
replaced_text = ' '.join(['*' * len(i) for i in text.split(' ')])
indices.append(list(range(index, index + token_length)))
index += token_length
all_text = all_text[:idx + text_len].replace(
text, replaced_text) + all_text[idx + text_len:]
for date_indices in indices:
date = ''
left, top, right, bottom = [], [], [], []
for i in date_indices:
date += ' ' + all_words[i]['text']
left.append(all_words[i]['left'])
top.append(all_words[i]['top'])
right.append(all_words[i]['left'] + all_words[i]['width'])
bottom.append(all_words[i]['top'] + all_words[i]['height'])
cad = {
'text': all_words[i]['text'],
'x1': all_words[i]['left'],
'y1': all_words[i]['top'],
'x2': all_words[i]['left'] + all_words[i]['width'],
'y2': all_words[i]['top'] + all_words[i]['height']
}
neighbours = Neighbour.find_neighbour(cad, words, int(width * 0.3),
int(height * 0.02), width,
height)
for item in neighbours:
if 'due' in item['text'].lower():
all_dates.append({
'text': date.strip(),
'x1': min(left),
'y1': min(top),
'x2': max(right),
'y2': max(bottom)
})
return all_dates