A comparison of computational color constancy algorithms--part I: methodology and experiments with synthesized data

Area of Science:

• Computer Vision
• Image Processing
• Computational Photography

Background:

• Accurate color representation in images is crucial for various applications.
• Computational color constancy aims to correct for variations in scene illumination.
• Existing methods often lack standardized evaluation contexts and ground truth data.

Purpose of the Study:

• To establish a controlled environment for evaluating computational color constancy algorithms.
• To implement and assess the performance of prominent color constancy algorithms.
• To investigate factors influencing algorithm accuracy using synthesized data.

Main Methods:

• Development of a synthesized data context for computational color constancy testing.
• Implementation of key algorithms: Gray World, Retinex, Gamut-Mapping, Neural Net, Color by Correlation.
• Experimental evaluation focusing on illuminant chromaticity, magnitude, and illumination-invariant images.

Main Results:

• Algorithm performance was analyzed based on scene complexity (number of surfaces).
• The impact of specularities and data clipping on algorithm accuracy was quantified.
• Synthesized data enabled precise investigation of individual factors affecting color constancy.

Conclusions:

• The proposed synthesized data context provides a robust framework for algorithm evaluation.
• Different algorithms exhibit varying sensitivities to scene properties and data manipulations.
• The study provides valuable insights into the strengths and weaknesses of current color constancy techniques.