def test_clamp():
assert utils.clamp(1, 50, 100) == 50
assert utils.clamp(1, 1, 100) == 1
assert utils.clamp(1, 100, 100) == 100
assert utils.clamp(1, -2, 100) == 1
assert utils.clamp(0, 1, 0) == 0
assert utils.clamp(1, 99, 100) == 99
assert utils.clamp(1, 110, 100) == 100
def _match_template(image: Image.Image, template: Image.Image,
tolerance: float) -> Iterator[Region]:
"""Use opencv's matchTemplate() to slide the `template` over
`image` to calculate correlation coefficients, and then
filter with a tolerance to find all relevant global maximums.
"""
template_width, template_height = template.size
if image.mode == "RGBA":
image = image.convert("RGB")
if template.mode == "RGBA":
template = template.convert("RGB")
image = numpy.array(image)
template = numpy.array(template)
image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
template = cv2.cvtColor(template, cv2.COLOR_RGB2BGR)
# Template matching result is a single channel array of shape:
# Width: Image width - template width + 1
# Height: Image height - template height + 1
coefficients = cv2.matchTemplate(image, template, cv2.TM_CCOEFF_NORMED)
coeff_height, coeff_width = coefficients.shape
while True:
# The point (match_x, match_y) is the top-left of the best match
_, match_coeff, _, (match_x, match_y) = cv2.minMaxLoc(coefficients)
if match_coeff < tolerance:
break
# Zero out values for a template-sized region around the best match
# to prevent duplicate matches for the same element.
left = clamp(0, match_x - template_width // 2, coeff_width)
top = clamp(0, match_y - template_height // 2, coeff_height)
right = clamp(0, match_x + template_width // 2, coeff_width)
bottom = clamp(0, match_y + template_height // 2, coeff_height)
coefficients[top:bottom, left:right] = 0
yield Region.from_size(match_x, match_y, template_width,
template_height)
def _to_tolerance(confidence):
"""Convert confidence value to tolerance.
Confidence is a logarithmic scale from 1 to 100,
tolerance is a linear scale from 0.01 to 1.00.
"""
value = float(confidence)
value = clamp(1, value, 100)
value = log2lin(1, value, 100)
value = value / 100.0
return value
def find(image: Union[Image.Image, Path],
text: str,
confidence: float = DEFAULT_CONFIDENCE):
"""Scan image for text and return a list of regions
that contain it (or something close to it).
:param image: Path to image or Image object
:param text: Text to find in image
:param confidence: Minimum confidence for text similaritys
"""
image = to_image(image)
confidence = clamp(1, float(confidence), 100)
text = str(text).strip()
if not text:
raise ValueError("Empty search string")
try:
data = pytesseract.image_to_data(image,
output_type=pytesseract.Output.DICT)
except TesseractNotFoundError as err:
raise EnvironmentError(INSTALL_PROMPT) from err
lines = defaultdict(list)
for word in _iter_rows(data):
if word["level"] != 5:
continue
if not word["text"].strip():
continue
key = "{:d}-{:d}-{:d}".format(word["block_num"], word["par_num"],
word["line_num"])
region = Region.from_size(word["left"], word["top"], word["width"],
word["height"])
# NOTE: Currently ignoring confidence in tesseract results
lines[key].append({"text": word["text"], "region": region})
assert len(lines[key]) == word["word_num"]
matches = _match_lines(lines.values(), text, confidence)
return matches
def find(
image: Union[Image.Image, Path],
text: str,
confidence: float = DEFAULT_CONFIDENCE,
region: Optional[Region] = None,
):
"""Scan image for text and return a list of regions
that contain it (or something close to it).
:param image: Path to image or Image object
:param text: Text to find in image
:param confidence: Minimum confidence for text similaritys
"""
image = to_image(image)
confidence = clamp(1, float(confidence), 100)
text = str(text).strip()
if not text:
raise ValueError("Empty search string")
if region is not None:
region = geometry.to_region(region)
image = image.crop(region.as_tuple())
try:
data = pytesseract.image_to_data(image, output_type=pytesseract.Output.DICT)
except TesseractNotFoundError as err:
raise EnvironmentError(INSTALL_PROMPT) from err
lines = _dict_lines(data)
matches = _match_lines(lines, text, confidence)
if region is not None:
for match in matches:
match["region"] = match["region"].move(region.left, region.top)
return matches