Simple Assumptions to Improve Markov Illuminance and Reflectance.

Area of Science:

• Computational neuroscience
• Computer vision
• Psychophysics

Background:

• The Markov illuminance and reflectance (MIR) model simulates human lightness processing using conditional random fields.
• Existing MIR models struggle with reverse-contrast phenomena, limiting their predictive accuracy for certain visual illusions.

Purpose of the Study:

• To enhance the MIR model's predictive capabilities by refining its inference process and priors.
• To improve the model's accuracy in simulating human lightness perception, particularly for challenging visual effects.

Main Methods:

• Modified the inference process within the MIR framework.
• Adjusted priors for X-junctions and general illumination changes.
• Evaluated the enhanced model against various lightness illusions, including Checkerboard assimilation, Checkershadow, luminance noise, and White's effect.

Main Results:

• The modified MIR model demonstrated improved predictions for Checkerboard assimilation, Checkershadow, luminance noise, and White's effect variants.
• Achieved significant progress in predicting partial reverse-contrast phenomena, such as White's effect, a known challenge for computational models.
• Validated the enhanced model's ability to account for a wider range of lightness illusions and perceptual phenomena.

Conclusions:

• The enhanced MIR model shows significant improvements in predicting human lightness perception.
• The modifications highlight the MIR framework's extensibility and potential for further development in computational vision.
• This work advances computational modeling of visual perception, offering a more robust tool for understanding lightness illusions.