SSR, MSR, and Custom Improved Algorithms Based on Retinex Theory

This discussion covers Single Scale Retinex (SSR) and Multi-Scale Retinex (MSR) algorithms derived from Retinex theory, supplemented by our custom improvements. Retinex theory represents a color restoration and image enhancement framework designed to address color distortion and illumination inconsistencies in digital images. The SSR algorithm operates through Gaussian filtering to estimate illumination components, followed by logarithmic domain operations to extract reflectance properties. The MSR algorithm enhances this approach by combining multiple Gaussian scales (typically 3 scales with varying sigma values) to preserve both local contrast and global tonal balance. Our improvements include optimized scale selection strategies and parallel computation implementations using OpenCV's GaussianBlur() function with kernel size automation. We've also integrated adaptive histogram equalization techniques to handle high dynamic range scenarios. These algorithmic refinements demonstrate significant improvements in processing efficiency (reducing runtime by approximately 30% through multithreading) while maintaining perceptual naturalness. We believe these advancements contribute meaningfully to image processing research and enable innovative applications in computational photography and computer vision systems.