I Need help with detecting and identifying rectangular gums in images using a rule-based image processing approach and overlaying detected gums with green and red rectangles based on shape.
what’s there to “identify” ?
why green or red rects? (based on what ?)
then, please show current effort, ty.
i have provided the sample image. i have more images like it to identify
i have images where gums are not rectangular or irregular in shape .so the green will represent valid and red one will represent defective
noone here will “write you a program”, we can only try to improve your attempt, so, what’s currently blocking you ?
import cv2
import numpy as np
import matplotlib.pyplot as plt
# Load the image
img_path = "/content/ANAIMAGE_151B.BMP" # Update with the correct path
img = cv2.imread(img_path)
if img is None:
raise ValueError("Image not loaded. Check the file path.")
img_copy = img.copy()
img_defects = img.copy()
# Convert image to grayscale and HSV for better analysis
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
# Define color range for gums (adjust if needed)
lower_gum = np.array([15, 20, 60])
upper_gum = np.array([35, 170, 255])
# Create mask for gum detection
mask = cv2.inRange(hsv, lower_gum, upper_gum)
# Morphological operations to remove noise
kernel = np.ones((5, 5), np.uint8)
mask = cv2.morphologyEx(mask, cv2.MORPH_CLOSE, kernel)
mask = cv2.morphologyEx(mask, cv2.MORPH_OPEN, kernel)
# Find contours
contours, _ = cv2.findContours(mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
# Get image dimensions
height, width = img.shape[:2]
row_height = height // 3 # Assuming 3 rows
# Store detected properties
valid_contours = []
for cnt in contours:
x, y, w, h = cv2.boundingRect(cnt)
if y < 2 * row_height and w > 20 and h > 10: # Only consider the first two rows
area = cv2.contourArea(cnt)
aspect_ratio = w / float(h) if h > 0 else 0
rectangularity = area / (w * h) if w * h > 0 else 0
# Define defect criteria
is_defect = rectangularity < 0.85 or aspect_ratio < 0.4 or aspect_ratio > 0.65
# Draw appropriate rectangles
if is_defect:
cv2.rectangle(img_defects, (x, y), (x + w, y + h), (0, 0, 255), 2) # Red for defective
else:
cv2.rectangle(img_defects, (x, y), (x + w, y + h), (0, 255, 0), 2) # Green for good
# Display results
plt.figure(figsize=(16, 8)) # Increased figure size
plt.subplot(1, 2, 1)
plt.title('Original Image with Detections')
plt.imshow(cv2.cvtColor(img_copy, cv2.COLOR_BGR2RGB))
plt.axis('off')
plt.subplot(1, 2, 2)
plt.title('Defect Detection')
plt.imshow(cv2.cvtColor(img_defects, cv2.COLOR_BGR2RGB))
plt.axis('off')
plt.show()
this is my current approach it is showing some false positives . it should process only the middle two rows
its me only. i have posted it on stackoverflow but i didnt get any response so i posted it here
you did receive two comments by one user on SO.
isn’t that wrong ? for a valid rect, area should be as close to w*h as possible. no ?
right, which is why less “rectangularity” is worse, and under 0.85 is defective.
1 Like
Next time put an image with good and defective gums first, currently your post starts with an image of good gums only, but no question or any info. Then later after showing your current approach you posted some low-res screenshot of some better picture that’s still not very usefull. Your image on stackoverflow is much better than all images you posted here.
I don’t see any false positives, only false negatives and they seem to occur at the left and right border of the image, because it’s a bit darker there. If you can, improve your lighting or try to find better parameters for your pre-processing. But when improving the lighting, keep in mind that reflections of your your dotted background make it also more complicated to seperate the gums from the rest;
Try to display your mask using imshow to see where your problems are. I don’t know about python but there’s a function called drawContours that you can use to paint them to img_defects, use -1 as contourIdx. Your left and right borders look very frayed.
I also don’t think that contourArea is very accurate. In C I would measure the area with something like area=countNonZero(mask(bbox)); where bbox is the boundingRect of your contour. Or draw your contours filled (thickness=-1) to an empty image and count the nonZero pixels there to remove the holes.
also check putText to draw some text into your images like “area” or “rectangularity”. Your false negatives have a ratio of ~0.83 and that’s why it fails. But doing a better pre-processing and improving the mask is a better way than decreasing this threshold.