really inefficient macocr, but hey it works (ecff441c) · Commits · git-mirror / BallonsTranslator

modules/ocr/init.py

+33 −18

Original line number	Diff line number	Diff line
		@@ -246,34 +246,37 @@ class OCRMIT48pxCTC(OCRBase):
		self.chunk_size = chunk_size
		self.model.max_chunk_size = chunk_size

		import sys, platform
		if sys.platform == 'darwin' and platform.mac_ver()[0] >= '10.15':
		import platform
		if platform.mac_ver()[0] >= '10.15':
		from .macos_ocr import get_supported_languages
		langs = list(get_supported_languages()[0])
		APPLEVISIONFRAMEWORK = None
		@register_OCR('macos_ocr')
		class OCRApple(OCRBase):
		params = {
		'Language': {
		'language': {
		'type':'selector',
		'options': langs
		'options': list(get_supported_languages()[0]),
		'select': 'en-US',
		},
		# 'Recognition Level': {
		# 'recognition_level': {
		# 'type': 'selector',
		# 'options': [
		# 'Fast',
		# 'Accurate'
		# ]
		# }
		# 'accurate',
		# 'fast',
		# ],
		# 'select': 'accurate',
		# },
		'confidence_level': '0.1',
		}
		language = 'en-US'
		# recognition = 'Accurate'
		recognition = 'accurate'
		confidence = '0.1'

		def setup_ocr(self):
		global APPLEVISIONFRAMEWORK
		from .macos_ocr import AppleOCR
		if APPLEVISIONFRAMEWORK is None:
		self.model = APPLEVISIONFRAMEWORK = AppleOCR()
		self.model = APPLEVISIONFRAMEWORK = AppleOCR(lang=[self.language])
		else:
		self.model = APPLEVISIONFRAMEWORK

		@@ -281,17 +284,29 @@ if sys.platform == 'darwin' and platform.mac_ver()[0] >= '10.15':
		return self.model(img)

		def ocr_blk_list(self, img: np.ndarray, blk_list: List[TextBlock]):
		pass
		im_h, im_w = img.shape[:2]
		for blk in blk_list:
		x1, y1, x2, y2 = blk.xyxy
		if y2 < im_h and x2 < im_w and \
		x1 > 0 and y1 > 0 and x1 < x2 and y1 < y2:
		blk.text = self.model(img[y1:y2, x1:x2])
		else:
		logging.warning('invalid textbbox to target img')
		blk.text = ['']

		def updateParam(self, param_key: str, param_content):
		super().updateParam(param_key, param_content)
		language = self.params['language']['select']
		# recognition = self.params['recognition level']['select']
		self.language = self.params['language']['select']
		self.model.lang = [self.language]

		# self.recognition = self.params['recognition_level']['select']
		# self.model.recog_level = self.recognition
		# self.params['language']['options'] = list(get_supported_languages(self.recognition)[0])

		self.language = language
		# self.recognition = recognition
		self.confidence = self.params['confidence_level']
		self.model.min_confidence = self.confidence

		if sys.platform == 'win32':
		if platform.system() == 'Windows':
		WINDOWSOCRENGINE = None
		@register_OCR('WindowsOCR')
		class OCRWindows(OCRBase):

modules/ocr/macos_ocr.py

+30 −24

Original line number	Diff line number	Diff line
		# https://github.com/straussmaximilian/ocrmac/blob/main/ocrmac/ocrmac.py
		# https://gist.github.com/RhetTbull/1c34fc07c95733642cffcd1ac587fc4c
		# https://github.com/RhetTbull/textinator/blob/main/src/macvision.py

		import Vision
		import objc
		import platform
		from typing import Tuple
		import numpy as np

		# Vision.VNRequestTextRecognitionLevelAccurate 0
		# Vision.VNRequestTextRecognitionLevelFast 1
		# Vision.VNRecognizeTextRequestRevision1 1
		# Vision.VNRecognizeTextRequestRevision2 2
		# Vision.VNRecognizeTextRequestRevision3 3
		from PIL import Image
		from io import BytesIO

		def get_revision_level():
		with objc.autorelease_pool():
		@@ -26,25 +18,32 @@ def get_revision_level():
		revision = Vision.VNRecognizeTextRequestRevision1
		return revision

		def get_supported_languages(recognition_level=0, revision=get_revision_level()) -> Tuple[Tuple[str], Tuple[str]]:
		def get_supported_languages(recognition_level='accurate', revision=get_revision_level()) -> Tuple[Tuple[str], Tuple[str]]:
		"""Get supported languages for text detection from Vision framework.

		Returns: Tuple of ((language code), (error))
		"""

		if recognition_level == 'fast':
		recognition_level = 1
		else:
		recognition_level = 0
		return Vision.VNRecognizeTextRequest.supportedRecognitionLanguagesForTextRecognitionLevel_revision_error_(
		recognition_level, revision, None
		)

		def text_from_image(image: np.ndarray, recognition_level="accurate", language_preference=None):
		def text_from_image(image: np.ndarray, language_preference=None, recognition_level='accurate'):
		recognition_level = recognition_level.lower()
		if language_preference == 'Auto':
		language_preference = None
		image = image.tobytes()

		img_buf = BytesIO()
		Image.fromarray(image).save(img_buf, format='PNG')

		with objc.autorelease_pool():
		req = Vision.VNRecognizeTextRequest.alloc().init()

		if recognition_level == "fast":
		if recognition_level == 'fast':
		req.setRecognitionLevel_(1)
		else:
		req.setRecognitionLevel_(0)
		@@ -53,18 +52,18 @@ def text_from_image(image: np.ndarray, recognition_level="accurate", language_pr
		req.setRecognitionLanguages_(language_preference)

		handler = Vision.VNImageRequestHandler.alloc().initWithData_options_(
		image, None
		img_buf.getvalue(), None
		)

		success = handler.performRequests_error_([req], None)
		res = []
		if success:
		for result in req.results():
		bbox = result.boundingBox()
		w, h = bbox.size.width, bbox.size.height
		x, y = bbox.origin.x, bbox.origin.y
		# bbox = result.boundingBox()
		# w, h = bbox.size.width, bbox.size.height
		# x, y = bbox.origin.x, bbox.origin.y

		res.append((result.text(), result.confidence(), [x, y, w, h]))
		res.append((result.text(), result.confidence())) #, [x, y, w, h]))

		req.dealloc()
		handler.dealloc()
		@@ -73,8 +72,15 @@ def text_from_image(image: np.ndarray, recognition_level="accurate", language_pr


		class AppleOCR:
		def __init__(self):
		pass

		def __call__(self, img) -> str:
		pass
		No newline at end of file
		def __init__(self, lang=[], recog_level='accurate', min_confidence='0.1'):
		self.lang = lang
		self.recog_level = recog_level
		self.min_confidence = min_confidence

		def __call__(self, img: np.ndarray) -> str:
		result = []
		results = text_from_image(img, self.lang, self.recog_level)
		for res in results:
		if res[1] >= float(self.min_confidence):
		result.append(res[0])
		return '\n'.join(result)
		No newline at end of file