Computational-observer analysis of illumination discrimination.

Area of Science:

• Visual perception
• Computational neuroscience
• Color science

Background:

• Ambient illumination's spectral properties encode environmental information like time and weather.
• Human visual system's ability to perceive illumination is crucial for scene understanding.
• Previous models often simplify the complex interplay between light, surfaces, and perception.

Purpose of the Study:

• To investigate the human perceptual representation of illumination.
• To compare human performance in discriminating illuminations with a computational observer model.
• To identify discrepancies between human perception and a biophysically-based computational model.

Main Methods:

• Human observers performed illumination discrimination tasks across various scene configurations.
• A computational observer model simulated cone photopigment isomerizations in a photoreceptor mosaic.
• Model performance was adjusted for noise to match human sensitivity levels for comparative analysis.

Main Results:

• Computational observer's baseline sensitivity exceeded human performance.
• Adding noise to the computational observer matched overall human performance levels.
• Specific human performance patterns, like blue-direction threshold elevations and surface ensemble effects, were not replicated by the model.

Conclusions:

• Cone isomerization noise alone does not fully account for human illumination discrimination.
• Human visual perception of illumination involves mechanisms beyond the information captured by cone isomerizations.
• Further research is needed to elucidate the neural processes underlying human illumination perception.