含七段数码的无文本图像角度检测与旋转校正的Python实现方案咨询

Hey there! I totally get your frustration—HoughLines works great for text because of its continuous horizontal/vertical strokes, but seven-segment digits have broken lines that throw that method off. And OCR is a non-starter with blurry images. Let's try a contour-based approach that leverages the distinct shape of seven-segment displays instead.

Core Idea #

Seven-segment digits have a consistent rectangular-like footprint, even when rotated. We can use the minimum enclosing rectangle of the digit contours to find the exact rotation angle, then correct the image accordingly. This doesn't rely on full lines or text recognition, so it handles blurriness much better.

Step-by-Step Implementation (Python + OpenCV) #

First, make sure you have OpenCV installed: pip install opencv-python numpy

1. Load and Preprocess the Image #

We'll convert the image to grayscale and apply thresholding to get a binary image where the segments are isolated from the background. This helps cut through noise, especially in blurry images.

import cv2
import numpy as np

# Load your target image
img = cv2.imread('your_seven_segment_image.jpg')
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

# Optional: Reduce noise for blurry images
gray = cv2.GaussianBlur(gray, (3, 3), 0)

# Create binary image (auto-adjust threshold with Otsu's method)
_, binary = cv2.threshold(gray, 127, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)
# For uneven lighting, swap to adaptive threshold:
# binary = cv2.adaptiveThreshold(gray, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY_INV, 11, 2)

2. Isolate Digit Contours #

We'll find all contours in the binary image and filter out small, noisy ones to focus solely on the seven-segment digits.

# Find only external contours (outer edges of digits)
contours, _ = cv2.findContours(binary, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)

# Filter contours by size (adjust min area based on your image resolution)
min_contour_area = 500  # Tweak this value for your specific images
digit_contours = [cnt for cnt in contours if cv2.contourArea(cnt) > min_contour_area]

# Get a combined bounding box for all digits (assuming they're grouped together)
if digit_contours:
    combined_contour = np.vstack(digit_contours)
    x, y, w, h = cv2.boundingRect(combined_contour)
    digit_roi = binary[y:y+h, x:x+w]
else:
    # Fallback: use the entire image if no valid contours are detected
    digit_roi = binary

3. Calculate Rotation Angle #

The minimum enclosing rectangle of the digit region will give us the exact angle we need to rotate the image to align the digits properly. We'll adjust the angle to ensure we rotate in the correct direction.

# Get the minimum enclosing rectangle of the digit region
rect = cv2.minAreaRect(digit_roi)
angle = rect[2]

# Adjust angle for correct rotation (OpenCV returns angles between -90 and 0)
if angle < -45:
    angle = 90 + angle

# Now angle is in the range [-45, 45], which is the amount to rotate to align digits

4. Rotate the Image #

Finally, we'll apply the rotation to the original image, using a solid border to fill in any gaps created by the rotation.

# Get image center and generate rotation matrix
(h, w) = img.shape[:2]
center = (w // 2, h // 2)
rot_matrix = cv2.getRotationMatrix2D(center, angle, 1.0)

# Rotate the image (use INTER_CUBIC for smooth resizing)
rotated_img = cv2.warpAffine(
    img, 
    rot_matrix, 
    (w, h), 
    flags=cv2.INTER_CUBIC, 
    borderMode=cv2.BORDER_CONSTANT, 
    borderValue=(0, 0, 0)  # Fill gaps with black background
)

# Save or display the corrected image
cv2.imwrite('corrected_image.jpg', rotated_img)
cv2.imshow('Corrected Image', rotated_img)
cv2.waitKey(0)
cv2.destroyAllWindows()

Key Tips for Success #

• Tweak Parameters: Adjust the contour area threshold, blur kernel size, and threshold values based on your specific images. What works for one might not work for another.
• Handle Spread-Out Digits: If your digits are spaced apart, calculate the average angle of each digit's minimum enclosing rectangle instead of using a combined contour.
• Strengthen Contours: For very blurry images, add morphological operations like erosion/dilation to thicken weak segment lines before contour detection.

备注：内容来源于stack exchange，提问作者sadiq ali