Three Methods Based on Retinex Theory: SSR, MSR, and Self-Quotient Image

This implementation provides three distinct Retinex-based image enhancement approaches: Single-Scale Retinex (SSR) for basic illumination compensation using Gaussian surround functions, Multi-Scale Retinex (MSR) that combines multiple Gaussian scales for robust dynamic range compression, and Self-Quotient Image method for illumination-invariant representation through pixel-wise division operations. The code architecture employs modular design with separate directories for each method, containing core functions for logarithmic transformation, convolution operations with Gaussian kernels, and image normalization routines. Each directory includes a test script that automatically loads sample images, applies the respective Retinex algorithm, and displays before/after comparisons. The implementation features configurable parameters including Gaussian sigma values for scale control in SSR/MSR, and threshold settings for quotient image normalization. Comprehensive documentation details the mathematical foundations of each method, including the Retinex equation implementation (I(x,y) = L(x,y) * R(x,y)) and practical considerations for parameter tuning. Users can modify the source code to adjust convolution kernel sizes, experiment with different normalization techniques, or integrate custom image preprocessing pipelines. The package includes sample images for validation and performance benchmarking against standard datasets. For technical support or implementation queries, please refer to the included documentation or contact the development team through the provided channels.