def extract_image_data(data, languages=None):
"""Extract text from a binary string of data."""
tessdata_prefix = get_config('TESSDATA_PREFIX')
if tessdata_prefix is None:
raise IngestorException("TESSDATA_PREFIX is not set, OCR won't work.")
languages = get_languages_iso3(languages)
text = Cache.get_ocr(data, languages)
if text is not None:
return text
try:
img = Image.open(StringIO(data))
except DecompressionBombWarning as dce:
log.debug("Image too large: %", dce)
return None
except IOError as ioe:
log.info("Unknown image format: %r", ioe)
return None
# TODO: play with contrast and sharpening the images.
extractor = Tesseract(tessdata_prefix, lang=languages)
extractor.set_image(img)
extractor.set_page_seg_mode(PageSegMode.PSM_AUTO_OSD)
text = extractor.get_text() or ''
text = text.decode(encoding="UTF-8")
# extractor.clear()
log.debug('OCR done: %s, %s characters extracted', languages, len(text))
Cache.set_ocr(data, languages, text)
return text
def ocr_text(img):
'''Perform OCR on the image.'''
tr = Tesseract(lang='eng')
tr.clear()
pil_image = pil.Image.fromarray(img)
tr.set_image(pil_image)
utf8_text = tr.get_text()
return utf8_text
def ocr(img,idioma):
ocr_img = Image.fromarray(img)
ocr = Tesseract(lang=idioma)
ocr.set_image(ocr_img)
pattern = re.compile('[a-zA-Z0-9]')
text = ocr.get_utf8_text()
text = text.splitlines()
text = [x for x in text if x != '']
text = [x for x in text if pattern.search(x)]
ocr.clear()
return (text)
def ocrImage(tagDest,tessdataPrefix,lang,charWhitelist,pageMode):
destOcrImg = "/tmp/"+genymotion_vm_name+"-"+tagDest+".png"
print "OCR : "+str(destOcrImg)
#OCR Def
tr = Tesseract(tessdataPrefix, lang)
tr.set_variable("tessedit_char_whitelist", charWhitelist)
tr.set_page_seg_mode(pageMode)
#OCR
image = Image.open(destOcrImg)
tr.set_image(image)
return tr.get_utf8_text()
def ocr_text(img):
tr = Tesseract(lang='eng')
tr.clear()
pil_image = pil.Image.fromarray(img)
# Turn off OCR word dictionaries
tr.set_variable('load_system_dawg', "F")
tr.set_variable('load_freq_dawg', "F")
tr.set_variable('-psm', "7") # treat image as single line
tr.set_variable('tessedit_char_whitelist', "ABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789")
tr.set_image(pil_image)
utf8_text = tr.get_text()
return unicode(utf8_text)
class PaperDetection:
def __init__(self):
cwd = os.path.dirname(os.path.realpath(__file__))
os.environ["TESSDATA_PREFIX"] = cwd
self.tr = Tesseract(lang="deu")
self.gs = goslate.Goslate()
self.trained_paper = False
self.paper_row_nw = None
self.paper_row_se = None
self.paper_col_nw = None
self.paper_col_se = None
self.paper_hist = None
self.paper = None
self.words = None
self.translations = []
self.pointed_locations = deque(maxlen=20)
def draw_paper_rect(self, frame):
rows, cols, _ = frame.shape
self.paper_row_nw = rows / 5
self.paper_row_se = 4 * rows / 5
self.paper_col_nw = 2 * cols / 5
self.paper_col_se = 3 * cols / 5
cv2.rectangle(
frame, (self.paper_col_nw, self.paper_row_nw), (self.paper_col_se, self.paper_row_se), (0, 255, 0), 1
)
black = np.zeros(frame.shape, dtype=frame.dtype)
frame_final = np.vstack([frame, black])
return frame_final
def train_paper(self, frame):
self.set_paper_hist(frame)
self.trained_paper = True
def get_paper(self, frame):
paper_masked = image_analysis.apply_hist_mask(frame, self.paper_hist)
contours = image_analysis.contours(paper_masked)
max_contour = image_analysis.max_contour(contours)
paper = image_analysis.contour_interior(frame, max_contour)
return paper
def set_paper(self, frame):
self.paper = self.get_paper(frame)
def paper_copy(self):
paper = self.paper.copy()
return paper
def set_ocr_text(self, frame):
paper = self.get_paper(frame)
thresh = image_analysis.gray_threshold(paper, 100)
paper_img = Image.fromarray(thresh)
self.tr.set_image(paper_img)
self.tr.get_text()
self.words = self.tr.get_words()
for w in self.words:
translation = self.translate(w.value)
self.translations.append(translation)
def set_paper_hist(self, frame):
hsv = cv2.cvtColor(frame, cv2.COLOR_BGR2HSV)
roi = hsv[self.paper_row_nw : self.paper_row_se, self.paper_col_nw : self.paper_col_se]
self.paper_hist = cv2.calcHist([roi], [0, 1], None, [180, 256], [0, 180, 0, 256])
cv2.normalize(self.paper_hist, self.paper_hist, 0, 255, cv2.NORM_MINMAX)
def get_word_at_point(self, point):
for i, w in enumerate(self.words):
x_nw, y_nw, x_se, y_sw = w.box
x, y = point
if x > x_nw and x < x_se and y > y_nw and y < y_sw:
return self.translations[i]
def get_word_index(self, point):
for i, w in enumerate(self.words):
x_nw, y_nw, x_se, y_sw = w.box
x, y = point
if x > x_nw and x < x_se and y > y_nw and y < y_sw:
return i
def translate(self, word):
translated_word = self.gs.translate(word, "en", source_language="de")
return translated_word
def update_pointed_locations(self, point):
index = self.get_word_index(point)
if index != None:
self.pointed_locations.append(index)
def get_most_common_word(self):
index = self.most_common_location()
if index != None:
word = self.translations[index].encode("ascii", errors="backslashreplace")
return word
def most_common_location(self):
values = set(self.pointed_locations)
index = None
maxi = 0
for i in values:
num = self.pointed_locations.count(i)
if num > maxi:
index = i
frequency = float(self.pointed_locations.count(index)) / float(self.pointed_locations.maxlen)
if frequency > 0.25:
return index
else:
return None
class PaperDetection:
def __init__(self):
cwd = os.path.dirname(os.path.realpath(__file__))
os.environ['TESSDATA_PREFIX'] = cwd
self.tr = Tesseract(lang='deu')
self.gs = goslate.Goslate()
self.trained_paper = False
self.paper_row_nw = None
self.paper_row_se = None
self.paper_col_nw = None
self.paper_col_se = None
self.paper_hist = None
self.paper = None
self.words = None
self.translations = []
self.pointed_locations = deque(maxlen=20)
def draw_paper_rect(self, frame):
rows,cols,_ = frame.shape
self.paper_row_nw = rows/5
self.paper_row_se = 4*rows/5
self.paper_col_nw = 2*cols/5
self.paper_col_se = 3*cols/5
cv2.rectangle(frame,(self.paper_col_nw,self.paper_row_nw),(self.paper_col_se,self.paper_row_se),
(0,255,0),1)
black = np.zeros(frame.shape, dtype=frame.dtype)
frame_final = np.vstack([frame, black])
return frame_final
def train_paper(self, frame):
self.set_paper_hist(frame)
self.trained_paper = True
def get_paper(self, frame):
paper_masked = image_analysis.apply_hist_mask(frame, self.paper_hist)
contours = image_analysis.contours(paper_masked)
max_contour = image_analysis.max_contour(contours)
paper = image_analysis.contour_interior(frame, max_contour)
return paper
def set_paper(self, frame):
self.paper = self.get_paper(frame)
def paper_copy(self):
paper = self.paper.copy()
return paper
def set_ocr_text(self, frame):
paper = self.get_paper(frame)
thresh = image_analysis.gray_threshold(paper, 100)
paper_img = Image.fromarray(thresh)
self.tr.set_image(paper_img)
self.tr.get_text()
self.words = self.tr.get_words()
for w in self.words:
translation = self.translate(w.value)
self.translations.append(translation)
def set_paper_hist(self, frame):
hsv = cv2.cvtColor(frame, cv2.COLOR_BGR2HSV)
roi = hsv[self.paper_row_nw:self.paper_row_se, self.paper_col_nw:self.paper_col_se]
self.paper_hist = cv2.calcHist([roi], [0, 1], None, [180, 256], [0, 180, 0, 256])
cv2.normalize(self.paper_hist, self.paper_hist, 0, 255, cv2.NORM_MINMAX)
def get_word_at_point(self, point):
for i, w in enumerate(self.words):
x_nw,y_nw,x_se,y_sw = w.box
x,y = point
if x > x_nw and x < x_se and y > y_nw and y < y_sw:
return self.translations[i]
def get_word_index(self, point):
for i, w in enumerate(self.words):
x_nw,y_nw,x_se,y_sw = w.box
x,y = point
if x > x_nw and x < x_se and y > y_nw and y < y_sw:
return i
def translate(self, word):
translated_word = self.gs.translate(word,'en',source_language='de')
return translated_word
def update_pointed_locations(self, point):
index = self.get_word_index(point)
if index != None:
self.pointed_locations.append(index)
def get_most_common_word(self):
index = self.most_common_location()
if index != None:
word = self.translations[index].encode('ascii', errors='backslashreplace')
return word
def most_common_location(self):
values = set(self.pointed_locations)
index = None
maxi = 0
for i in values:
num = self.pointed_locations.count(i)
if num > maxi:
index = i
frequency = float(self.pointed_locations.count(index))/float(self.pointed_locations.maxlen)
if frequency > 0.25:
return index
else:
return None
