def __init__(self, frame):
super(KeyboardRecognizer, self).__init__()
self.image = Image(frame)
proportions = reversed([(3. / 16, 7. / 16), (6. / 16, 3. / 16),
(10. / 16, 3. / 16), (13. / 16, 7. / 16)])
width = 400
height = int(400 / sqrt(2))
self.elatole_size = (width, height)
self.etalone_orients = [
Point(width * px, height * py) for px, py in proportions
]
def get_keyboard(self):
transformed_image = self.transformed()
if transformed_image is not None:
transformed_image = fix_channels(transformed_image)
gray = transformed_image.to_gray()
gray = gray.eroded((3, 3))
height, width = gray.get_size()
recog_height = int(0.6 * height)
return Image(gray.image[recog_height:, :])
return None
def __init__(self, frame):
super(KeyboardRecognizer, self).__init__()
self.image = Image(frame)
proportions = reversed(
[(3.0 / 16, 7.0 / 16), (6.0 / 16, 3.0 / 16), (10.0 / 16, 3.0 / 16), (13.0 / 16, 7.0 / 16)]
)
width = 400
height = int(400 / sqrt(2))
self.elatole_size = (width, height)
self.etalone_orients = [Point(width * px, height * py) for px, py in proportions]
class Recognizer(object):
""" Recognizes the keyboard image using its topology """
def __init__(self, frame):
super(Recognizer, self).__init__()
self.image = Image(frame)
proportions = ((3./16, 7./16),
(6./16, 3./16),
(10./16, 3./16),
(13./16, 7./16))
width = 400
height = int(400 / sqrt(2))
self.elatole_size = (width, height)
self.etalone_orients = [Point(width * px, height * py)
for px, py in proportions]
def get_all_ellipses(self):
gray = self.image.to_gray()
thresh = gray.get_treshold()
contours = thresh.find_contours()
ellipses = [cnt for cnt in contours if cnt.is_ellipse()]
return ellipses
def recognize(self):
''' Returns the tranformation matrix'''
ellipses = self.get_all_ellipses()
if len(ellipses) == 2:
return []
elif len(ellipses) == 1:
C = ellipses[0]
if C.get_centroid().x < self.image.image.shape[0] / 2:
return ['C-4']
else:
return ['F-5']
else:
return ['C-4', 'F-5']
def get_notes(self):
return self.recognize()
class Recognizer(object):
""" Recognizes the keyboard image using its topology """
def __init__(self, frame):
super(Recognizer, self).__init__()
self.image = Image(frame)
proportions = ((3. / 16, 7. / 16), (6. / 16, 3. / 16),
(10. / 16, 3. / 16), (13. / 16, 7. / 16))
width = 400
height = int(400 / sqrt(2))
self.elatole_size = (width, height)
self.etalone_orients = [
Point(width * px, height * py) for px, py in proportions
]
def get_all_ellipses(self):
gray = self.image.to_gray()
thresh = gray.get_treshold()
contours = thresh.find_contours()
ellipses = [cnt for cnt in contours if cnt.is_ellipse()]
return ellipses
def recognize(self):
''' Returns the tranformation matrix'''
ellipses = self.get_all_ellipses()
if len(ellipses) == 2:
return []
elif len(ellipses) == 1:
C = ellipses[0]
if C.get_centroid().x < self.image.image.shape[0] / 2:
return ['C-4']
else:
return ['F-5']
else:
return ['C-4', 'F-5']
def get_notes(self):
return self.recognize()
def fix_channels(image):
(R, G, B) = image.split()
return Image.from_channels([R, G, R])
class KeyboardRecognizer(object):
""" Recognizes the keyboard image using its topology """
def __init__(self, frame):
super(KeyboardRecognizer, self).__init__()
self.image = Image(frame)
proportions = reversed([(3. / 16, 7. / 16), (6. / 16, 3. / 16),
(10. / 16, 3. / 16), (13. / 16, 7. / 16)])
width = 400
height = int(400 / sqrt(2))
self.elatole_size = (width, height)
self.etalone_orients = [
Point(width * px, height * py) for px, py in proportions
]
def get_all_ellipses(self):
gray = self.image.to_gray()
thresh = gray.get_treshold()
contours = thresh.find_contours()
ellipses = [cnt for cnt in contours if cnt.is_ellipse()]
return ellipses
def choose_four(self, ellipses):
gray = self.image.to_gray()
ellipses = [
e for e in ellipses if gray.get_pixel(e.center.reversed()) < 100
] # with dark enters only
if len(ellipses) < 4:
return None
candidates = []
for four in combinations(ellipses, 4):
centers = [e.center for e in four]
cnt = Contour.from_points(centers)
cnt = cnt.make_convex()
if len(cnt) != 4:
continue
centers = cnt.to_points()
centers = sorted(centers, key=lambda p: p.x)
areas = [e.get_area() for e in four]
diff = max(areas) - min(areas)
candidates.append((diff, centers))
centers = min(candidates, key=lambda x: x[0])[1]
return centers
def transformed(self):
''' Returns the tranformed image'''
ellipses = self.get_all_ellipses()
for e in ellipses:
e.draw(self.image, color=(255, 60, 60))
orients = self.choose_four(ellipses)
if orients is not None:
cnt = Contour.from_points(orients)
cnt.draw(self.image, color=(60, 60, 255))
corrected_image = self.image.perspective(orients,
self.etalone_orients,
self.elatole_size)
return corrected_image
return None
def get_keyboard(self):
transformed_image = self.transformed()
if transformed_image is not None:
transformed_image = fix_channels(transformed_image)
gray = transformed_image.to_gray()
gray = gray.eroded((3, 3))
height, width = gray.get_size()
recog_height = int(0.6 * height)
return Image(gray.image[recog_height:, :])
return None
def fix_channels(image):
(R, G, B) = image.split()
return Image.from_channels([R, G, R])
class KeyboardRecognizer(object):
""" Recognizes the keyboard image using its topology """
def __init__(self, frame):
super(KeyboardRecognizer, self).__init__()
self.image = Image(frame)
proportions = reversed(
[(3.0 / 16, 7.0 / 16), (6.0 / 16, 3.0 / 16), (10.0 / 16, 3.0 / 16), (13.0 / 16, 7.0 / 16)]
)
width = 400
height = int(400 / sqrt(2))
self.elatole_size = (width, height)
self.etalone_orients = [Point(width * px, height * py) for px, py in proportions]
def get_all_ellipses(self):
gray = self.image.to_gray()
thresh = gray.get_treshold()
contours = thresh.find_contours()
ellipses = [cnt for cnt in contours if cnt.is_ellipse()]
return ellipses
def choose_four(self, ellipses):
gray = self.image.to_gray()
ellipses = [e for e in ellipses if gray.get_pixel(e.center.reversed()) < 100] # with dark enters only
if len(ellipses) < 4:
return None
candidates = []
for four in combinations(ellipses, 4):
centers = [e.center for e in four]
cnt = Contour.from_points(centers)
cnt = cnt.make_convex()
if len(cnt) != 4:
continue
centers = cnt.to_points()
centers = sorted(centers, key=lambda p: p.x)
areas = [e.get_area() for e in four]
diff = max(areas) - min(areas)
candidates.append((diff, centers))
centers = min(candidates, key=lambda x: x[0])[1]
return centers
def transformed(self):
""" Returns the tranformed image"""
ellipses = self.get_all_ellipses()
for e in ellipses:
e.draw(self.image, color=(255, 60, 60))
orients = self.choose_four(ellipses)
if orients is not None:
cnt = Contour.from_points(orients)
cnt.draw(self.image, color=(60, 60, 255))
corrected_image = self.image.perspective(orients, self.etalone_orients, self.elatole_size)
return corrected_image
return None
def get_keyboard(self):
transformed_image = self.transformed()
if transformed_image is not None:
transformed_image = fix_channels(transformed_image)
gray = transformed_image.to_gray()
gray = gray.eroded((3, 3))
height, width = gray.get_size()
recog_height = int(0.6 * height)
return Image(gray.image[recog_height:, :])
return None
