# Imports
from .common import Any, NDArray, check_image, cv2, np
# Functions
[docs]
def canny_image(
image: NDArray[Any],
threshold1: float,
threshold2: float,
aperture_size: int = 3,
sigma: float = 3,
stop_at_nms: bool = False,
ignore_dtype: bool = False,
) -> NDArray[Any]:
""" Apply Canny edge detection to an image.
Args:
image (NDArray[Any]): Image to apply Canny edge detection
threshold1 (float): First threshold for hysteresis (between 0 and 1)
threshold2 (float): Second threshold for hysteresis (between 0 and 1)
aperture_size (int): Aperture size for Sobel operator (3, 5, or 7)
sigma (float): Standard deviation for Gaussian blur
stop_at_nms (bool): Stop at non-maximum suppression step (don't apply thresholding)
ignore_dtype (bool): Ignore the dtype check
Returns:
NDArray[Any]: Image with Canny edge detection applied
>>> ## Basic tests
>>> image = np.array([[100, 150, 200], [50, 125, 175], [25, 75, 225]])
>>> edges = canny_image(image.astype(np.uint8), 0.1, 0.2)
>>> edges.shape == image.shape[:2] # Canny returns single channel
True
>>> set(np.unique(edges).tolist()) <= {0, 255} # Only contains 0 and 255
True
>>> rgb = np.random.randint(0, 256, (3,3,3), dtype=np.uint8)
>>> edges_rgb = canny_image(rgb, 0.1, 0.2)
>>> edges_rgb.shape == rgb.shape[:2] # Canny returns single channel
True
>>> ## Test invalid inputs
>>> canny_image("not an image", 0.1, 0.2)
Traceback (most recent call last):
...
AssertionError: Image must be a numpy array
>>> canny_image(image.astype(np.uint8), 1.5, 0.2)
Traceback (most recent call last):
...
AssertionError: threshold1 must be between 0 and 1, got 1.5
"""
# Check input data
check_image(image, ignore_dtype=ignore_dtype)
assert 0 <= threshold1 <= 1, f"threshold1 must be between 0 and 1, got {threshold1}"
assert 0 <= threshold2 <= 1, f"threshold2 must be between 0 and 1, got {threshold2}"
assert aperture_size in (3, 5, 7), f"aperture_size must be 3, 5 or 7, got {aperture_size}"
# Convert to grayscale if needed
if len(image.shape) > 2:
image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
# Convert thresholds to 0-255 range
t1: int = int(threshold1 * 255)
t2: int = int(threshold2 * 255)
if not stop_at_nms:
# Apply full Canny edge detection
return cv2.Canny(image, t1, t2, apertureSize=aperture_size)
else:
# Manual implementation up to non-maximum suppression
# 1. Apply Gaussian blur to reduce noise
blurred: NDArray[Any] = cv2.GaussianBlur(image, (5, 5), sigma)
# 2. Calculate gradients using Sobel
# Use constant value 6 for 64-bit float depth (CV_64F)
gx: NDArray[Any] = cv2.Sobel(blurred, 6, 1, 0, ksize=aperture_size)
gy: NDArray[Any] = cv2.Sobel(blurred, 6, 0, 1, ksize=aperture_size)
# Calculate gradient magnitude and direction
magnitude: NDArray[Any] = np.sqrt(gx**2 + gy**2)
direction: NDArray[Any] = np.arctan2(gy, gx) * 180 / np.pi
# 3. Non-maximum suppression
nms: NDArray[Any] = np.zeros_like(magnitude, dtype=np.uint8)
height, width = magnitude.shape
for i in range(1, height - 1):
for j in range(1, width - 1):
# Get gradient direction
angle: float = direction[i, j]
if angle < 0:
angle += 180
# Round to 0, 45, 90, or 135 degrees
if (0 <= angle < 22.5) or (157.5 <= angle <= 180):
neighbors = [magnitude[i, j-1], magnitude[i, j+1]]
elif 22.5 <= angle < 67.5:
neighbors = [magnitude[i-1, j-1], magnitude[i+1, j+1]]
elif 67.5 <= angle < 112.5:
neighbors = [magnitude[i-1, j], magnitude[i+1, j]]
else: # 112.5 <= angle < 157.5
neighbors = [magnitude[i-1, j+1], magnitude[i+1, j-1]]
# Check if current pixel is local maximum
if magnitude[i, j] >= max(neighbors):
nms[i, j] = np.uint8(min(magnitude[i, j], 255))
return nms
