Multi-Scale Retinex Algorithm

Multi-Scale Retinex (MSR) is a classical image enhancement algorithm that effectively improves uneven illumination, enhances shadow details, and maintains natural color reproduction. Its core concept originates from the color constancy characteristic of human visual system - the human eye's ability to recognize object colors under varying lighting conditions.

The algorithm implementation consists of three key steps: Multi-scale Gaussian filtering: The input image undergoes blurring processing at different scales (typically small, medium, and large scales), simulating the human eye's adaptability to lighting variations. Logarithmic reflectance component extraction: The logarithmic difference between the original image and each blurred result separates the reflectance component representing the object's intrinsic properties (the de-illuminated image). Weighted fusion: Linear combination of reflectance components from different scales balances local contrast enhancement with global tone balance.

MATLAB implementation highlights: Use `imfilter` with Gaussian kernels for efficient convolution Handle zero values before logarithmic operations to prevent numerical overflow Control scale contributions through weight adjustments (e.g., [0.25, 0.5, 0.25])

Typical applications include medical imaging, low-light surveillance, and remote sensing image processing, though excessive enhancement may amplify noise. Common improvements incorporate MSRCR with color restoration or noise suppression strategies.