Colour Constancy Beyond the Classical Receptive Field.

Area of Science:

• Computer Vision
• Computational Neuroscience
• Image Processing

Background:

• Colour constancy, preserving object colours under varying illumination, is crucial for visual perception.
• The human brain achieves colour constancy through centre-surround computations of local contrast.
• Existing computational models often lack the adaptability of biological systems.

Purpose of the Study:

• To develop a novel computational model for colour constancy inspired by human visual mechanisms.
• To create a fully automatic functional model named Adaptive Surround Modulation (ASM).
• To investigate the role of dynamical adaptation in achieving accurate colour constancy.

Main Methods:

• Developed the Adaptive Surround Modulation (ASM) model incorporating biological principles of receptive field size and surround modulation.
• Modeled centre-surround interactions using two overlapping asymmetric Gaussian kernels with contrast-adapted sizes.
• Simulated contrast-dependent surround modulation and estimated illuminants from activated receptive field outputs.

Main Results:

• The ASM model demonstrated highly competitive performance against state-of-the-art methods on benchmark datasets.
• ASM outperformed learning-based algorithms in one case, highlighting its effectiveness.
• Consistent results across multiple datasets using identical parameters underscore the model's robustness and biological plausibility.

Conclusions:

• Dynamical adaptation mechanisms are key to achieving high accuracy in computational colour constancy.
• The ASM model offers a promising biologically-inspired approach for robust colour constancy.
• This work suggests that mimicking human visual system's adaptive strategies can enhance artificial vision systems.