如何消除图像光照影响并优化OpenCV图像处理方案的效果

Looking at your current code, you’ve already built a solid foundation using background subtraction, normalization, and gamma correction to reduce lighting effects—but let’s tweak a few key parts to fix the dark central area and blurry top-right lines. Here are targeted optimizations tailored to your Python 3.6.5 + OpenCV 3.4.0 environment:

1. Improve Background Extraction to Preserve Edge Details #

Your current background uses dilation + median blur, which can over-blur fine edges (like the top-right lines). Swap the median blur for bilateral filtering—it smooths flat areas while retaining sharp edges, so your background will be more accurate without losing critical line details:

# Replace bg_img = cv2.medianBlur(dilated_img, 21) with:
bg_img = cv2.bilateralFilter(dilated_img, d=21, sigmaColor=75, sigmaSpace=75)

You can also try a slightly smaller dilation kernel (like (5,5) instead of (7,7)) if the current dilation is overly aggressive and smudges small details.

2. Replace Global Normalization with Adaptive Histogram Equalization #

The global cv2.NORM_MINMAX normalization suppresses local dark areas (like your central region) because it’s tied to the entire image’s brightness range. Switch to CLAHE (Contrast-Limited Adaptive Histogram Equalization) to enhance contrast locally:

# Remove the cv2.normalize call for diff_img, and replace with:
clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8,8))
norm_img = clahe.apply(diff_img)

This brightens the dark central area independently without washing out bright regions, and the clipLimit prevents over-amplifying noise in flat areas.

3. Adjust Thresholding and Gamma Correction Parameters #

• Truncation Threshold: Your current value of 253 is too high—it clips almost all bright pixels, including the top-right lines. Lower it to 240 for gentler truncation, or use adaptive thresholding for better local control:

  # Replace _, thr_img = cv2.threshold(norm_img, 253, 0, cv2.THRESH_TRUNC) with:
  thr_img = cv2.adaptiveThreshold(norm_img, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY, 11, 2)
  

• Gamma Correction: Your current gamma value (0.3) is quite aggressive, which amplifies noise in dark areas while not giving enough boost to the center. Try a slightly higher gamma like 0.4 for balanced brightness:

  table = np.array([((i / 255.0) ** (1.0/0.4)) * 255 for i in np.arange(0, 256)]).astype("uint8")
  

4. Add Minor Denoising to Clean Up Artifacts #

After computing the difference image, a quick Gaussian blur will reduce small noise artifacts that interfere with subsequent processing:

diff_img = 255 - cv2.absdiff(image, bg_img)
diff_img = cv2.GaussianBlur(diff_img, (3,3), 0)  # Add this line

Full Modified Code Example #

Here’s how the updated parse function looks with all these tweaks:

def parse(image):
    dilated_img = cv2.dilate(image, np.ones((5, 5), np.uint8))  # Slightly smaller dilation kernel
    bg_img = cv2.bilateralFilter(dilated_img, d=21, sigmaColor=75, sigmaSpace=75)  # Bilateral filter instead of median blur
    
    diff_img = 255 - cv2.absdiff(image, bg_img)
    diff_img = cv2.GaussianBlur(diff_img, (3,3), 0)  # Denoise
    if debug:
        cv2.imshow('origin vs back diff', diff_img)
        cv2.waitKey(0)
    
    # Use CLAHE instead of global normalization
    clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8,8))
    norm_img = clahe.apply(diff_img)
    if debug:
        cv2.imshow('CLAHE enhanced', norm_img)
        cv2.waitKey(0)
    
    # Adaptive thresholding to preserve top-right lines
    thr_img = cv2.adaptiveThreshold(norm_img, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY, 11, 2)
    
    # Adjusted gamma correction
    table = np.array([((i / 255.0) ** (1.0/0.4)) * 255 for i in np.arange(0, 256)]).astype("uint8")
    thr_img = cv2.LUT(thr_img, table)
    if debug:
        cv2.imshow('final processed', thr_img)
        cv2.waitKey(0)
    
    cv2.destroyAllWindows()
    # Fixed typo "reult.JPEG" to "result.JPEG" (adjust if your filename is different)
    compare = cv2.resize(np.hstack([image, cv2.imread("result.JPEG", 0), thr_img]), None, fx=0.5, fy=0.5)
    cv2.imshow("Analysis", compare)
    cv2.waitKey(0)
    cv2.destroyAllWindows()
    cv2.imwrite('./nijie.jpg', thr_img)

Key Notes #

• Bilateral filtering ensures the background doesn’t blur out fine lines in the top-right.
• CLAHE fixes the dark central region by enhancing contrast locally instead of globally.
• Adaptive thresholding adapts to local brightness, keeping top-right lines sharp even in brighter areas.
• The adjusted gamma value gives balanced brightness without over-amplifying noise.

内容的提问来源于stack exchange，提问作者Wesley