QBRIX: a quantile-based approach to retinex.

Area of Science:

• Computer Vision
• Image Processing
• Computational Photography

Background:

• Retinex algorithms are widely used for image enhancement by adjusting pixel lightness based on a reference white value.
• Existing Retinex methods employ diverse strategies for determining this reference white value, impacting spatial property analysis.
• A need exists for novel Retinex formalizations that explore alternative methods for reference white value computation.

Purpose of the Study:

• To introduce a novel probabilistic version of the Retinex algorithm.
• To investigate the spatial properties of this probabilistic Retinex model.
• To propose an alternative procedure for computing the reference white value using pixel population percentiles.

Main Methods:

• Probabilistic formalization of the random spray Retinex sampling.
• Development of two algorithm versions: global and local behavior.
• Computation of reference white values from percentile values of the pixel population.

Main Results:

• Introduction of a probabilistic Retinex model enhancing spatial property investigation.
• Successful implementation of percentile-based reference white value computation.
• Characterization of global and local Retinex versions with distinct computational costs.

Conclusions:

• The proposed probabilistic Retinex offers a new perspective on image processing and spatial analysis.
• Percentile-based reference white value calculation provides an effective alternative to existing methods.
• The developed global and local versions cater to different computational requirements in image analysis.