如何基于Python实现仅提亮图像暗部区域

I feel your pain—regular histogram equalization is great for overall contrast but totally ruins bright areas when you only want to fix the dark parts. Let's walk through a couple of Python-based approaches using OpenCV and NumPy that will target just the dark regions, keeping your highlights intact, just like your example where the girl's hair and facial shadows get lifted without washing out the rest.

Approach 1: Targeted Linear/Gamma Brightening (Simple & Fast) #

This method separates the image's brightness channel, masks out only the dark areas, and applies a brightness boost only to those regions. We'll use the YCbCr color space since it neatly separates luminance (brightness) from color information.

Code Implementation #

import cv2
import numpy as np

def brighten_dark_areas(image_path, dark_threshold=100, brightness_gain=1.8):
    # Load the image (OpenCV reads in BGR format by default)
    img = cv2.imread(image_path)
    if img is None:
        raise ValueError("Could not read the image. Check the file path.")
    
    # Convert to YCbCr to isolate the luminance channel
    ycbcr_img = cv2.cvtColor(img, cv2.COLOR_BGR2YCrCb)
    y_channel, cr_channel, cb_channel = cv2.split(ycbcr_img)
    
    # Create a mask for dark regions (where luminance is below our threshold)
    dark_mask = y_channel < dark_threshold
    
    # Convert luminance to float to avoid overflow during calculations
    y_float = y_channel.astype(np.float32)
    
    # Apply brightness gain only to dark regions
    y_float[dark_mask] *= brightness_gain
    
    # Ensure values stay within valid 0-255 range
    y_float = np.clip(y_float, 0, 255)
    brightened_y = y_float.astype(np.uint8)
    
    # Merge channels back and convert to BGR
    brightened_ycbcr = cv2.merge([brightened_y, cr_channel, cb_channel])
    final_img = cv2.cvtColor(brightened_ycbcr, cv2.COLOR_YCrCb2BGR)
    
    return final_img

Customization Tips #

• Adjust dark_threshold: Set this to the luminance value that separates "dark" from "bright" (e.g., 80 for very dark regions, 120 for moderately shadowed areas).
• Swap linear gain for gamma correction: If you want a more natural-looking brightening (instead of linear scaling), replace the gain line with:

  gamma = 0.5  # Values <1 brighten, >1 darken
  y_float[dark_mask] = 255 * ((y_float[dark_mask] / 255) ** gamma)
  

Approach 2: CLAHE for Dark Regions (Preserves Local Detail) #

If your dark regions have fine details (like hair texture), Contrast Limited Adaptive Histogram Equalization (CLAHE) is better—it adjusts contrast locally instead of globally. We'll apply CLAHE only to the masked dark regions to avoid affecting bright areas.

Code Implementation #

import cv2
import numpy as np

def clahe_brighten_dark_regions(image_path, dark_threshold=100, clip_limit=2.0, tile_grid_size=(8,8)):
    img = cv2.imread(image_path)
    if img is None:
        raise ValueError("Could not read the image. Check the file path.")
    
    # Isolate luminance channel via YCbCr
    ycbcr_img = cv2.cvtColor(img, cv2.COLOR_BGR2YCrCb)
    y_channel, cr_channel, cb_channel = cv2.split(ycbcr_img)
    
    dark_mask = y_channel < dark_threshold
    
    # Initialize CLAHE (controls contrast limit and local tile size)
    clahe = cv2.createCLAHE(clipLimit=clip_limit, tileGridSize=tile_grid_size)
    # Apply CLAHE to the entire luminance channel
    clahe_y = clahe.apply(y_channel)
    
    # Replace only dark regions with CLAHE-adjusted values; keep bright areas untouched
    brightened_y = np.where(dark_mask, clahe_y, y_channel)
    
    # Reconstruct final image
    brightened_ycbcr = cv2.merge([brightened_y, cr_channel, cb_channel])
    final_img = cv2.cvtColor(brightened_ycbcr, cv2.COLOR_YCrCb2BGR)
    
    return final_img

Customization Tips #

• clip_limit: Lower values (1.0-2.0) keep contrast natural; higher values (3.0+) create more dramatic contrast.
• tile_grid_size: Smaller grids (e.g., (4,4)) focus on finer details; larger grids ((16,16)) handle broader shadow areas.

How to Use #

Call either function with your image path, then save or display the result:

# Example usage
result = clahe_brighten_dark_regions("your_image.jpg", dark_threshold=90, clip_limit=1.5)
cv2.imwrite("brightened_result.jpg", result)
cv2.imshow("Original vs Result", np.hstack([cv2.imread("your_image.jpg"), result]))
cv2.waitKey(0)
cv2.destroyAllWindows()

This should give you the exact effect you're looking for—lifting dark regions like hair and facial shadows while preserving the original brightness of lighter areas.

内容的提问来源于stack exchange，提问作者arnino