Classification images for detection, contrast discrimination, and identification tasks with a common ideal observer

Area of Science:

• Visual neuroscience
• Perceptual psychology
• Computational vision

Background:

• The ideal observer model predicts a single strategy for visual tasks with identical signals.
• Human visual perception often deviates from ideal observer predictions.

Purpose of the Study:

• To investigate whether human observers employ consistent visual strategies for detection, contrast discrimination, and identification tasks.
• To analyze human visual strategies using classification image analysis and compare them to the ideal observer.

Main Methods:

• Forced-choice visual tasks (detection, contrast discrimination, identification) were presented in Gaussian white noise.
• Classification image analysis was used to evaluate human observers' visual strategies.
• The difference-of-Gaussians (DOG) profile was used as the signal for all tasks.

Main Results:

• Human observers exhibited significantly different classification images across the three tasks.
• Variability was prominent at low spatial frequencies (<5 cycles per degree), showing frequency enhancement (detection), suppression, and reversal (contrast discrimination).
• Classification images for the identification task closely matched the ideal observer filter.

Conclusions:

• Human visual strategies are task-dependent, diverging from the ideal observer model, especially in detection and contrast discrimination.
• Nonlinear transducers and spatial uncertainty may explain deviations from ideal observer performance.
• Understanding these differences is crucial for models of human visual perception.