### MSRCR Algorithm Implementation in Python The Multi-Scale Retinex with Color Restoration (MSRCR) algorithm enhances images by improving their visual quality while preserving or restoring natural colors. Below is an example of how this algorithm can be implemented using Python, leveraging libraries such as NumPy and OpenCV. #### Required Libraries To implement the MSRCR algorithm in Python, you need to install the following libraries: - `numpy`: For numerical operations. - `opencv-python`: For image processing tasks like reading/writing images and applying filters. You can install these packages via pip: ```bash pip install numpy opencv-python ``` #### Code Example for MSRCR Algorithm Below is a complete implementation of the MSRCR algorithm in Python: ```python import cv2 import numpy as np def msrcr(image, scales, G=50, b=25, alpha=125, beta=46, gain=1.0): """ Implements the Multi-Scale Retinex with Color Restoration (MSRCR). Parameters: image: Input RGB image array. scales: List of scale values for multi-scale retinex computation. G: Parameter controlling intensity scaling. b: Bias parameter. alpha: Scaling factor for color restoration term. beta: Scaling factor for local contrast enhancement term. gain: Gain adjustment factor. Returns: Enhanced image after applying MSRCR. """ img = image.astype(np.float32) img /= 255.0 # Compute log chromaticity channels L = np.log1p(cv2.cvtColor(img, cv2.COLOR_RGB2GRAY)) R = [] for s in scales: blurred = cv2.GaussianBlur(L, (s * 2 + 1, s * 2 + 1), 0) R.append(np.exp(blurred - L)) # Combine results from different scales M = sum(R) / len(scales) # Apply color restoration C = alpha * (np.log(alpha * img + 1)) ** beta # Final output combining both terms result = gain * (C * M + b) / (G + M) # Normalize back to [0, 255] result *= 255.0 result = np.clip(result, 0, 255).astype(np.uint8) return result if __name__ == "__main__": input_image_path = "input.jpg" output_image_path = "output_msrcr.jpg" # Read the input image original_img = cv2.imread(input_image_path) if original_img is None: raise ValueError("Image not found!") # Convert BGR to RGB since OpenCV reads images in BGR format rgb_img = cv2.cvtColor(original_img, cv2.COLOR_BGR2RGB) # Define parameters for MSRCR scales = [15, 80, 200] # Perform MSRCR enhanced_img = msrcr(rgb_img, scales=scales) # Save the processed image cv2.imwrite(output_image_path, cv2.cvtColor(enhanced_img, cv2.COLOR_RGB2BGR)) ``` This code defines a function called `msrcr` which applies the MSRCR algorithm to enhance the given image. The key steps include computing logarithmic luminance (`L`) and performing Gaussian blurring at various scales before combining them into one final representation. Additionally, it incorporates color restoration through specific mathematical transformations defined within the algorithm's framework[^2]. --- ### Explanation of Key Components 1. **Logarithmic Transformation**: This step converts pixel intensities into logarithmic space, facilitating better handling of dynamic range differences between light and dark regions[^3]. 2. **Multi-Scale Blurring**: By applying Gaussian blur at varying kernel sizes specified by `scales`, we capture details across multiple spatial frequencies, enhancing global illumination adjustments without losing fine textures. 3. **Color Restoration Term**: To ensure that enhanced images retain realistic hues despite increased brightness/contrast levels, a separate formula adjusts each channel based on its relative contribution compared to others[^4]. 4. **Normalization & Clipping**: After completing computations involving floating-point arithmetic, normalization ensures outputs remain valid unsigned integers suitable for storage/display purposes[^1]. --- ### Performance Considerations While effective, implementing advanced algorithms like MSRCR requires careful tuning depending upon target applications' requirements regarding speed versus accuracy tradeoffs. Moreover, selecting appropriate hyperparameters significantly impacts outcomes; thus experimentation may prove necessary when adapting generic implementations towards specialized use cases.