def ocr(img, config=''):
"""Runs Tesseract on a given image. Writes an intermediate tempfile and then runs the tesseract command on the image.
This is a simplified modification of image_to_string from PyTesseract, which is adapted to SKImage rather than PIL.
In principle we could have reimplemented it just as well - there are some apparent bugs in PyTesseract (e.g. it
may lose the NamedTemporaryFile due to its auto-delete behaviour).
:param mrz_mode: when this is True (default) the tesseract is configured to recognize MRZs rather than arbitrary texts.
"""
input_file_name = '%s.bmp' % pytesseract.tempnam()
output_file_name_base = '%s' % pytesseract.tempnam()
output_file_name = "%s.txt" % output_file_name_base
try:
imsave(input_file_name, img)
status, error_string = pytesseract.run_tesseract(input_file_name,
output_file_name_base,
lang=None,
boxes=False,
config=config)
if status:
errors = pytesseract.get_errors(error_string)
raise pytesseract.TesseractError(status, errors)
if 'vie' in config:
f = codecs.open(output_file_name, encoding='utf-8')
else:
f = open(output_file_name)
try:
return f.read().strip()
finally:
f.close()
finally:
pytesseract.cleanup(input_file_name)
pytesseract.cleanup(output_file_name)
def ocr(img, mrz_mode=True, extra_cmdline_params=''):
"""Runs Tesseract on a given image. Writes an intermediate tempfile and then runs the tesseract command on the image.
This is a simplified modification of image_to_string from PyTesseract, which is adapted to SKImage rather than PIL.
In principle we could have reimplemented it just as well - there are some apparent bugs in PyTesseract, but it works so far :)
:param mrz_mode: when this is True (default) the tesseract is configured to recognize MRZs rather than arbitrary texts.
When False, no specific configuration parameters are passed (and you are free to provide your own via `extra_cmdline_params`)
:param extra_cmdline_params: extra parameters passed to tesseract. When mrz_mode=True, these are appended to whatever is the
"best known" configuration at the moment.
"--oem 0" is the parameter you might want to pass. This selects the Tesseract's "legacy" OCR engine, which often seems
to work better than the new LSTM-based one.
"""
if img is None or img.shape[-1] == 0: # Issue #34
return ''
input_file_name = '%s.bmp' % _tempnam()
output_file_name_base = '%s' % _tempnam()
output_file_name = "%s.txt" % output_file_name_base
try:
# Prevent annoying warning about lossy conversion to uint8
if str(img.dtype).startswith(
'float') and np.nanmin(img) >= 0 and np.nanmax(img) <= 1:
img = img.astype(np.float64) * (np.power(2.0, 8) - 1) + 0.499999999
img = img.astype(np.uint8)
imwrite(input_file_name, img)
if mrz_mode:
# NB: Tesseract 4.0 does not seem to support tessedit_char_whitelist
config = (
"--psm 6 -c tessedit_char_whitelist=ABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789><"
" -c load_system_dawg=F -c load_freq_dawg=F {}"
).format(extra_cmdline_params)
else:
config = "{}".format(extra_cmdline_params)
pytesseract.run_tesseract(input_file_name,
output_file_name_base,
'txt',
lang=None,
config=config)
if sys.version_info.major == 3:
f = open(output_file_name, encoding='utf-8')
else:
f = open(output_file_name)
try:
return f.read().strip()
finally:
f.close()
finally:
pytesseract.cleanup(input_file_name)
pytesseract.cleanup(output_file_name)
def ocr(img, mrz_mode=True, extra_cmdline_params=''):
"""Runs Tesseract on a given image. Writes an intermediate tempfile and then runs the tesseract command on the image.
This is a simplified modification of image_to_string from PyTesseract, which is adapted to SKImage rather than PIL.
In principle we could have reimplemented it just as well - there are some apparent bugs in PyTesseract, but it works so far :)
:param mrz_mode: when this is True (default) the tesseract is configured to recognize MRZs rather than arbitrary texts.
:param extra_cmdline_params:extra parameters to the ocr.py
"""
input_file_name = '%s.bmp' % _tempnam()
output_file_name_base = '%s' % _tempnam()
output_file_name = "%s.txt" % output_file_name_base
try:
imsave(input_file_name, img)
if mrz_mode:
# NB: Tesseract 4.0 does not seem to support tessedit_char_whitelist
config = "--psm 6 -c tessedit_char_whitelist=ABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789>< -c load_system_dawg=F -c load_freq_dawg=F {}".format(
extra_cmdline_params)
else:
config = ""
pytesseract.run_tesseract(input_file_name,
output_file_name_base,
'txt',
lang=None,
config=config)
if sys.version_info.major == 3:
f = open(output_file_name, encoding='utf-8')
else:
f = open(output_file_name)
try:
return f.read().strip()
finally:
f.close()
finally:
pytesseract.cleanup(input_file_name)
pytesseract.cleanup(output_file_name)
def ocr(img, mrz_mode=True):
"""Runs Tesseract on a given image. Writes an intermediate tempfile and then runs the tesseract command on the image.
This is a simplified modification of image_to_string from PyTesseract, which is adapted to SKImage rather than PIL.
In principle we could have reimplemented it just as well - there are some apparent bugs in PyTesseract (e.g. it
may lose the NamedTemporaryFile due to its auto-delete behaviour).
:param mrz_mode: when this is True (default) the tesseract is configured to recognize MRZs rather than arbitrary texts.
"""
input_file_name = '%s.bmp' % pytesseract.tempnam()
output_file_name_base = '%s' % pytesseract.tempnam()
output_file_name = "%s.txt" % output_file_name_base
try:
imsave(input_file_name, img)
if mrz_mode:
config = "-psm 6 -c tessedit_char_whitelist=ABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789>< -c load_system_dawg=F -c load_freq_dawg=F"
else:
config = None
status, error_string = pytesseract.run_tesseract(input_file_name,
output_file_name_base,
lang=None,
boxes=False,
config=config)
if status:
errors = pytesseract.get_errors(error_string)
raise pytesseract.TesseractError(status, errors)
f = open(output_file_name)
try:
return f.read().strip()
finally:
f.close()
finally:
pytesseract.cleanup(input_file_name)
pytesseract.cleanup(output_file_name)
