Absolutely, there's a practical, step-by-step workflow to estimate the percentage of a specific colored structure in an image using OpenCV. Let's walk through this using your red eye vessel detection use case—since you've already tried masks and transforms, we'll refine those techniques to get better results.

Step 1: Preprocess the Image to Reduce Noise #

First, smooth the image to eliminate small noise that can throw off your mask. Gaussian blur is ideal here because it preserves edges while reducing high-frequency noise:

import cv2
import numpy as np

# Load your image (OpenCV reads in BGR by default)
img = cv2.imread("red_eye_image.jpg")
# Apply Gaussian blur
blurred = cv2.GaussianBlur(img, (5, 5), 0)

Step 2: Switch to a Color Space Optimized for Color Detection #

RGB isn't the best for isolating specific colors because it mixes brightness and color information. Instead, use HSV (Hue-Saturation-Value)—the hue channel directly represents color, making it easier to target red.

Red has two hue ranges in HSV: ~0-10 and ~170-180 (since hue wraps around the color wheel). Define your lower and upper bounds for red:

# Convert to HSV
hsv = cv2.cvtColor(blurred, cv2.COLOR_BGR2HSV)

# Define red color ranges (adjust these based on your specific image!)
lower_red1 = np.array([0, 50, 50])
upper_red1 = np.array([10, 255, 255])
lower_red2 = np.array([170, 50, 50])
upper_red2 = np.array([180, 255, 255])

# Combine the two masks to capture all red areas
mask1 = cv2.inRange(hsv, lower_red1, upper_red1)
mask2 = cv2.inRange(hsv, lower_red2, upper_red2)
mask = cv2.bitwise_or(mask1, mask2)

Step 3: Refine the Mask with Morphological Operations #

Your initial mask might have small noise spots or tiny gaps in the vessel structures. Use morphological operations to clean this up:

• Opening (erosion followed by dilation): Removes small white noise spots
• Closing (dilation followed by erosion): Fills small gaps in the vessel regions

# Create a kernel for morphological operations (adjust size based on noise)
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))

# Apply opening to remove noise
mask_cleaned = cv2.morphologyEx(mask, cv2.MORPH_OPEN, kernel, iterations=1)
# Apply closing to fill gaps
mask_cleaned = cv2.morphologyEx(mask_cleaned, cv2.MORPH_CLOSE, kernel, iterations=1)

Step 4: Calculate the Percentage of Vessel Pixels #

Now count the number of white pixels (vessels) in the mask and divide by the total number of pixels in the original image:

# Count non-zero (white) pixels in the mask
vessel_pixels = cv2.countNonZero(mask_cleaned)
# Total pixels in the image
total_pixels = img.shape[0] * img.shape[1]
# Calculate percentage
vessel_percentage = (vessel_pixels / total_pixels) * 100

print(f"Vessel percentage: {vessel_percentage:.2f}%")

Pro Tips for Better Results #

• Adjust Color Thresholds: Use cv2.imshow() to test different HSV bounds—you can even create a trackbar to tweak values in real-time for precise targeting.
• Enhance Contrast with CLAHE: If your image has low contrast (common in medical images like red eye shots), apply CLAHE to the luminance channel before processing:

  # Convert to LAB color space
  lab = cv2.cvtColor(blurred, cv2.COLOR_BGR2LAB)
  l, a, b = cv2.split(lab)
  # Apply CLAHE to the L channel
  clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8,8))
  l_clahe = clahe.apply(l)
  # Merge back and convert to HSV
  lab_clahe = cv2.merge((l_clahe, a, b))
  hsv = cv2.cvtColor(lab_clahe, cv2.COLOR_LAB2HSV)
  

• Filter Contours: If other red regions (like the red pupil) are being included, use contour detection to filter out areas that are too large or small:

  contours, _ = cv2.findContours(mask_cleaned, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
  # Create an empty mask to draw valid contours
  filtered_mask = np.zeros_like(mask_cleaned)
  # Only keep contours within a specific area range (adjust based on your vessel size)
  for cnt in contours:
      area = cv2.contourArea(cnt)
      if 10 < area < 500:  # Example bounds—tweak for your image
          cv2.drawContours(filtered_mask, [cnt], -1, 255, thickness=cv2.FILLED)
  # Use filtered_mask for percentage calculation instead
  

内容的提问来源于stack exchange，提问作者Joe Smith