[Paper Review] Entropy-guided Retinex anisotropic diffusion algorithm based on partial differential equations (PDE) for illumination correction

Uche A. Nnolim|arXiv (Cornell University)|Feb 4, 2017

Image Enhancement Techniques20 references3 citations

TL;DR

This paper proposes an entropy-guided Retinex anisotropic diffusion algorithm using partial differential equations (PDEs) for automatic illumination correction in images. By leveraging HSI/HSV color space and optimizing via perceptual quality metric stabilization and entropy maximization, the method enhances contrast, suppresses noise, eliminates color distortion and halos, and automatically determines stopping time—outperforming conventional Retinex methods in detail preservation and visual quality.

ABSTRACT

This report describes the experimental results obtained using a proposed variational Retinex algorithm for controlled illumination correction. Two colour restoration and enhancement schemes of the algorithm are presented for drastically improved results. The algorithm modifies the reflectance image using global and local contrast enhancement approaches and gradually removes the residual illumination to yield highly pleasing results. The proposed algorithms are optimized by way of simultaneous perceptual quality metric (PQM) stabilization and entropy maximization for fully automated processing solving the problem of determination of stopping time. The usage of the HSI or HSV colour space ensures a unique solution to the optimization problem unlike in the RGB space where there is none (forcing manual selection of number of iteration. The proposed approach preserves and enhances details in both bright and dark regions of underexposed images in addition to eliminating the colour distortion, over-exposure in bright image regions, halo effect and grey-world violations observed in Retinex-based approaches. Extensive experiments indicate consistent performance as the proposed approach exploits and augments the advantages of PDE-based formulation, performing illumination correction, colour enhancement correction and restoration, contrast enhancement and noise suppression. Comparisons shows that the proposed approach surpasses most of the other conventional algorithms found in the literature.

Motivation & Objective

• To address the challenge of automatic illumination correction in underexposed images with minimal user intervention.
• To overcome limitations of traditional Retinex methods, such as color distortion, halos, and over-exposure in bright regions.
• To eliminate the need for manual iteration count by introducing an automated stopping criterion.
• To enhance both dark and bright regions while preserving fine details and suppressing noise.
• To achieve a unique and stable solution through HSI/HSV color space transformation, avoiding ambiguities present in RGB space.

Proposed method

• The algorithm employs a variational Retinex framework based on partial differential equations (PDEs) for illumination correction.
• It uses HSI or HSV color space to decouple intensity (illumination) from chromaticity (reflectance), ensuring a unique optimization solution.
• Global and local contrast enhancement techniques are applied to the reflectance image to improve visual quality.
• Entropy maximization is used as a stopping criterion to automate the diffusion process and prevent over-smoothing.
• Simultaneous stabilization of the perceptual quality metric (PQM) ensures high-fidelity output without manual tuning.
• Anisotropic diffusion is guided by entropy and PQM to preserve edges and suppress noise while enhancing contrast.

Experimental results

Research questions

• RQ1 How can automatic stopping time be achieved in PDE-based illumination correction without manual iteration selection?
• RQ2 Can entropy maximization effectively guide the diffusion process to avoid over-smoothing and preserve image details?
• RQ3 How does the use of HSI/HSV color space improve solution uniqueness compared to RGB in Retinex-based methods?
• RQ4 To what extent does the proposed method suppress color distortion, halos, and over-exposure compared to conventional Retinex approaches?
• RQ5 Can the combination of PQM stabilization and entropy maximization yield consistent, high-quality results across diverse underexposed images?

Key findings

• The proposed method achieves automatic stopping time through simultaneous entropy maximization and perceptual quality metric stabilization, eliminating the need for manual iteration count.
• Extensive experiments show consistent performance across diverse images, with superior results in contrast enhancement and noise suppression.
• The algorithm effectively removes color distortion and prevents over-exposure in bright regions, which are common issues in traditional Retinex methods.
• Halo effects—typical in Retinex-based approaches—are significantly reduced due to the PDE-based anisotropic diffusion framework.
• The use of HSI/HSV color space ensures a unique solution, avoiding the ambiguity and manual tuning required in RGB-based Retinex implementations.
• Quantitative comparisons demonstrate that the proposed approach surpasses most conventional algorithms in the literature in visual quality and robustness.