from typing import List, Tuple
from .detect import _detect_text, _list_det_models
from .recognize import _text_recognize, _list_rec_models
from ..data import ImageTyping, load_image
_DEFAULT_DET_MODEL = 'ch_PP-OCRv4_det'
_DEFAULT_REC_MODEL = 'ch_PP-OCRv4_rec'
[docs]def list_det_models() -> List[str]:
"""
List available text detection models for OCR.
:return: A list of available text detection model names.
:rtype: List[str]
Examples::
>>> from imgutils.ocr import list_det_models
>>>
>>> list_det_models()
['ch_PP-OCRv2_det',
'ch_PP-OCRv3_det',
'ch_PP-OCRv4_det',
'ch_PP-OCRv4_server_det',
'ch_ppocr_mobile_slim_v2.0_det',
'ch_ppocr_mobile_v2.0_det',
'ch_ppocr_server_v2.0_det',
'en_PP-OCRv3_det']
"""
return _list_det_models()
[docs]def list_rec_models() -> List[str]:
"""
List available text recognition models for OCR.
:return: A list of available text recognition model names.
:rtype: List[str]
Examples::
>>> from imgutils.ocr import list_rec_models
>>>
>>> list_rec_models()
['arabic_PP-OCRv3_rec',
'ch_PP-OCRv2_rec',
'ch_PP-OCRv3_rec',
'ch_PP-OCRv4_rec',
'ch_PP-OCRv4_server_rec',
'ch_ppocr_mobile_v2.0_rec',
'ch_ppocr_server_v2.0_rec',
'chinese_cht_PP-OCRv3_rec',
'cyrillic_PP-OCRv3_rec',
'devanagari_PP-OCRv3_rec',
'en_PP-OCRv3_rec',
'en_PP-OCRv4_rec',
'en_number_mobile_v2.0_rec',
'japan_PP-OCRv3_rec',
'ka_PP-OCRv3_rec',
'korean_PP-OCRv3_rec',
'latin_PP-OCRv3_rec',
'ta_PP-OCRv3_rec',
'te_PP-OCRv3_rec']
"""
return _list_rec_models()
[docs]def detect_text_with_ocr(image: ImageTyping, model: str = _DEFAULT_DET_MODEL,
heat_threshold: float = 0.3, box_threshold: float = 0.7,
max_candidates: int = 1000, unclip_ratio: float = 2.0) \
-> List[Tuple[Tuple[int, int, int, int], str, float]]:
"""
Detect text in an image using an OCR model.
:param image: The input image.
:type image: ImageTyping
:param model: The name of the text detection model.
:type model: str, optional
:param heat_threshold: The heat map threshold for text detection.
:type heat_threshold: float, optional
:param box_threshold: The box threshold for text detection.
:type box_threshold: float, optional
:param max_candidates: The maximum number of candidates to consider.
:type max_candidates: int, optional
:param unclip_ratio: The unclip ratio for text detection.
:type unclip_ratio: float, optional
:return: A list of detected text boxes, label (always ``text``), and their confidence scores.
:rtype: List[Tuple[Tuple[int, int, int, int], str, float]]
Examples::
>>> from imgutils.ocr import detect_text_with_ocr
>>>
>>> detect_text_with_ocr('comic.jpg')
[((742, 485, 809, 511), 'text', 0.9543377610144915),
((682, 98, 734, 124), 'text', 0.9309689495575223),
((716, 136, 836, 164), 'text', 0.9042856988923695),
((144, 455, 196, 485), 'text', 0.874083638387722),
((719, 455, 835, 488), 'text', 0.8628696346175078),
((124, 478, 214, 508), 'text', 0.848871771901487),
((1030, 557, 1184, 578), 'text', 0.8352495440618789),
((427, 129, 553, 154), 'text', 0.8249209443996619)]
.. note::
If you need to extract the actual text content, use the :func:`ocr` function.
"""
retval = []
for box, _, score in _detect_text(image, model, heat_threshold, box_threshold, max_candidates, unclip_ratio):
retval.append((box, 'text', score))
retval = sorted(retval, key=lambda x: x[2], reverse=True)
return retval
[docs]def ocr(image: ImageTyping, detect_model: str = _DEFAULT_DET_MODEL,
recognize_model: str = _DEFAULT_REC_MODEL, heat_threshold: float = 0.3, box_threshold: float = 0.7,
max_candidates: int = 1000, unclip_ratio: float = 2.0, rotation_threshold: float = 1.5,
is_remove_duplicate: bool = False):
"""
Perform optical character recognition (OCR) on an image.
:param image: The input image.
:type image: ImageTyping
:param detect_model: The name of the text detection model.
:type detect_model: str, optional
:param recognize_model: The name of the text recognition model.
:type recognize_model: str, optional
:param heat_threshold: The heat map threshold for text detection.
:type heat_threshold: float, optional
:param box_threshold: The box threshold for text detection.
:type box_threshold: float, optional
:param max_candidates: The maximum number of candidates to consider.
:type max_candidates: int, optional
:param unclip_ratio: The unclip ratio for text detection.
:type unclip_ratio: float, optional
:param rotation_threshold: The rotation threshold for text detection.
:type rotation_threshold: float, optional
:param is_remove_duplicate: Whether to remove duplicate text content.
:type is_remove_duplicate: bool, optional
:return: A list of detected text boxes, their corresponding text content, and their combined confidence scores.
:rtype: List[Tuple[Tuple[int, int, int, int], str, float]]
Examples::
>>> from imgutils.ocr import ocr
>>>
>>> ocr('comic.jpg')
[((742, 485, 809, 511), 'MOB.', 0.9356705927336156),
((716, 136, 836, 164), 'SHISHOU,', 0.8933000384412466),
((682, 98, 734, 124), 'BUT', 0.8730931912907247),
((144, 455, 196, 485), 'OH,', 0.8417627579351514),
((427, 129, 553, 154), 'A MIRROR.', 0.7366019454049503),
((1030, 557, 1184, 578), '(EL) GATO IBERICO', 0.7271127306351021),
((719, 455, 835, 488), "THAt'S △", 0.701928390168364),
((124, 478, 214, 508), 'LOOK!', 0.6965972578194936)]
By default, the text recognition model used is `ch_PP-OCRv4_rec`.
This recognition model has good recognition capabilities for both Chinese and English.
For unsupported text types, its recognition accuracy cannot be guaranteed, resulting in a lower score.
**If you need recognition for other languages, please use :func:`list_rec_models` to
view more available recognition models and choose the appropriate one for recognition.**
>>> from imgutils.ocr import ocr
>>>
>>> # use default recognition model on japanese post
>>> ocr('post_text.jpg')
[
((319, 847, 561, 899), 'KanColle', 0.9130667787597329),
((552, 811, 791, 921), '1944', 0.8566762346615406),
((319, 820, 558, 850), 'Fleet Girls Collection', 0.8100635458911772),
((235, 904, 855, 1009), '海', 0.6716076803280185),
((239, 768, 858, 808), 'I · tSu · ka · A· NO· u· mI · de', 0.654507230718228),
((209, 507, 899, 811), '[', 0.2888084133529467)
]
>>>
>>> # use japanese model
>>> ocr('post_text.jpg', recognize_model='japan_PP-OCRv3_rec')
[
((319, 847, 561, 899), 'KanColle', 0.9230690942939336),
((552, 811, 791, 921), '1944', 0.8564870717047623),
((235, 904, 855, 1009), 'いつかあの海で', 0.8061289060358996),
((319, 820, 558, 850), 'Fleet Girls Collection', 0.8045396777081609),
((239, 768, 858, 808), 'I.TSU.KA・A・NO.U・MI.DE', 0.7311649382696896),
((209, 507, 899, 811), '「艦とれれ', 0.6648729016512889)
]
"""
image = load_image(image)
retval = []
for (x0, y0, x1, y1), _, score in \
_detect_text(image, detect_model, heat_threshold, box_threshold, max_candidates, unclip_ratio):
width, height = x1 - x0, y1 - y0
area = image.crop((x0, y0, x1, y1))
if height >= width * rotation_threshold:
area = area.rotate(90)
text, rec_score = _text_recognize(area, recognize_model, is_remove_duplicate)
retval.append(((x0, y0, x1, y1), text, score * rec_score))
retval = sorted(retval, key=lambda x: x[2], reverse=True)
return retval