Identification of illuminant and object colors: heuristic-based algorithms

Area of Science:

• Computer Vision
• Color Science
• Computational Neuroscience

Background:

• Human observers can identify objects and light sources (illuminants) from color information.
• Identifying objects and illuminants from sensor data is an underdetermined problem.
• Chromaticities of objects under different illuminants exhibit approximate affine transformations.

Purpose of the Study:

• To simplify the problem of object and illuminant identification from color information.
• To develop algorithms for identifying objects and illuminants using the affine property of chromaticities.
• To investigate the advantage of extracting object and illuminant information from retinal signals.

Main Methods:

• Utilizing the empirical result that chromaticities transform affinely across different illuminants.
• Developing algorithms that leverage this affine property as a heuristic.
• Estimating relative illuminant chromaticities during computation.

Main Results:

• Algorithms can identify objects with identical spectral reflectance across scenes with different illuminants.
• The method successfully estimates relative illuminant chromaticities.
• The approach simplifies the underdetermined problem of joint object and illuminant identification.

Conclusions:

• The affine transformation property provides a viable heuristic for solving the object and illuminant identification problem.
• Extracting both object and illuminant information from retinal signals is advantageous for the visual system.
• This method offers a more efficient approach than automatic discounting of information at early neural stages.