Multi-scale_FE (Implementation of "A Fusion-based Enhancing Method for Weakly Illuminated Images")

Python implementation of the algorithm in the paper "A Fusion-based Enhancing Method for Weakly Illuminated Images".

---

Usage

a. Go to the project directory.

cd Multi-scale_FE/

b. Generate illumination enhanced image.

python multi_scale_fe.py --input_path ${INPUT_PATH} (optional: --level ${LEVEL})

Please replace ${INPUT_PATH} and ${LEVEL} in the shell of the command line with the real input image path and the number of layers.

---

Optimizations

The multi_scale_fe.py script has undergone several optimizations to improve its performance and efficiency:

• box_filter Optimization: The original cumulative sum-based box_filter was replaced with cv2.boxFilter(..., normalize=False, borderType=cv2.BORDER_REPLICATE), leveraging OpenCV's highly optimized implementation for sum-based box filtering.
• guided_filter Efficiency: The calculation of the window area (N) within the guided_filter was changed from a box_filter call on an array of ones to a direct scalar computation (N = (2 * rad + 1)**2), saving a full image filtering operation.
• OpenCV Pyramid Functions: The manual Gaussian pyramid construction functions (pyramid_reduce and pyramid_expand) were replaced with OpenCV's optimized cv2.pyrDown and cv2.pyrUp functions, respectively. The generate_pyramid function was updated to use these standard functions, ensuring correct size handling for Laplacian pyramid construction.
• Vectorized Structuring Element: The loop-based generation of the structuring element (struct_elem) in fusion_based_method was replaced with direct NumPy array assignments for improved efficiency.
• Redundancy Reduction: Unnecessary data type conversions (e.g., redundant np.double() calls on already float64 arrays) were removed in several parts of the code, particularly in pyramid generation and image resizing loops.
• General Python Cleanups:
  • Unused imports (math, scipy.signal) were removed.
  • math.pi was replaced with np.pi.
  • Non-essential matplotlib.pyplot.show() calls were commented out to make the script suitable for non-interactive execution, relying on plt.imsave() for output.

These optimizations contribute to a more streamlined, faster, and more efficient execution of the image enhancement process.