On the RAID dataset of perceptual responses: analysis and statistical causes

TL;DR

This study analyzes the RAID dataset to understand human responses to various image distortions, revealing that noise sensitivity correlates with high-frequency energy and image likelihood influences perception.

Contribution

The paper provides a detailed statistical and spectral analysis of human perceptual thresholds for affine image distortions using the RAID dataset, introducing insights into visual masking and likelihood effects.

Findings

Humans are more sensitive to Gaussian noise than other distortions.

High-frequency components act as visual masks for Gaussian noise.

Image probability correlates with detection thresholds for most distortions.

Abstract

This work analyzes the RAID dataset to evaluate human responses to affine image distortions, including rotation, translation, scaling, and Gaussian noise. Using Mean Squared Error (MSE), the study establishes human detection thresholds for these distortions, enabling comparison across types. Statistical analysis with ANOVA and Tukey Kramer tests reveals that observers are significantly more sensitive to Gaussian noise, which consistently produced the lowest detection thresholds. Fourier analysis further shows that high-frequency components act as a visual mask for Gaussian noise, demonstrating a strong correlation between high frequency energy and detection thresholds. Additionally, spectral orientation influences the perception of rotation. Finally, the study employs the PixelCNN model to show that image probability significantly correlates with detection thresholds for most distortions, suggesting that statistical likelihood affects human visual tolerance.