### Retinex Algorithm Code Implementation The Retinex theory was initially proposed by Edwin H. Land to explain how humans perceive color under different lighting conditions[^1]. The algorithm aims at enhancing the visual quality of images, especially when dealing with non-uniform illumination. Below is a basic implementation of Single Scale Retinex (SSR) and Multi-Scale Retinex (MSR) algorithms using Python: #### SSR Implementation Single Scale Retinex enhances an image based on one scale parameter. ```python import numpy as np from scipy.ndimage import gaussian_filter import cv2 def single_scale_retinex(img, sigma): retinex_img = np.log10(img) - np.log10(cv2.GaussianBlur(img, (0, 0), sigma)) return retinex_img # Example usage: img = cv2.imread('input_image.jpg', cv2.IMREAD_GRAYSCALE).astype(np.float32) / 255.0 ssr_result = single_scale_retinex(img, sigma=3) cv2.imshow('SSR Result', ssr_result) cv2.waitKey(0) ``` This function applies Gaussian blurring followed by logarithmic operations to enhance local contrast within the specified scale defined by `sigma`[^2]. #### MSR Implementation Multi-Scale Retinex combines results from multiple scales for better enhancement. ```python def multi_scale_retinex(img, sigmas): msr_img = np.zeros_like(img) for sigma in sigmas: msr_img += single_scale_retinex(img, sigma) msr_img = msr_img / len(sigmas) return msr_img sigmas = [15, 80, 200] msr_result = multi_scale_retinex(img, sigmas=sigmas) cv2.imshow('MSR Result', msr_result) cv2.waitKey(0) ``` In this case, three distinct values are used for `sigmas`, which represent varying levels of smoothing applied during processing[^3]. For further improvements such as Color Restoration or dynamic range compression techniques can be integrated into these fundamental implementations depending upon specific requirements.