Retinexbased

Retinexbased is a class of image processing methods developed from the Retinex theory of color vision, used to enhance images and achieve color constancy under varying illumination. The Retinex concept was introduced to model an observed image I(x) as the product of a reflectance R(x) and an illumination L(x).

Many Retinex-based algorithms operate in the log domain, where log I = log R + log L, and estimate log L through a center-surround filter or a multiscale Gaussian surround. The objective is to recover an estimate of the reflectance R, which tends to be more independent of lighting.

Common variants include single-scale Retinex, multi-scale Retinex (MSR), and Retinex with Color Restoration (RRC). A widely used combination is MSR with Color Restoration (MSRCR), which adds a color-restoration function and dynamic range compression to produce more natural colours and contrast.

Applications include enhancement of under- and over-illuminated images, remote sensing, medical imaging, and consumer photography. Retinex-based methods are valued for improving local contrast and reducing shading effects while preserving edges.

Limitations include potential halo artifacts near edges, color shifts in some scenes, and sensitivity to parameter choices and computation time. Extensions often address color preservation, edge handling, or real-time implementation.