Exemplo n.º 1
0
    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
        ]
Exemplo n.º 2
0
    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
Exemplo n.º 3
0
    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]
Exemplo n.º 4
0
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()
Exemplo n.º 5
0
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()
Exemplo n.º 6
0
def fix_channels(image):
    (R, G, B) = image.split()
    return Image.from_channels([R, G, R])
Exemplo n.º 7
0
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
Exemplo n.º 8
0
def fix_channels(image):
    (R, G, B) = image.split()
    return Image.from_channels([R, G, R])
Exemplo n.º 9
0
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