Masking of Gaussian Noise in Color Images

A Psychophysical Study of Just-Noticeable Differences Using Synthetic Image Patches of Different Luminance Value

Understanding Noise Perception in Color Images

We are pleased to share that the IMaLeVICS team has published a new paper in the IbPRIA 2025 proceedings, titled “Masking of Gaussian Noise in Color Images: A Psychophysical Study of Just-Noticeable Differences Using Synthetic Image Patches of Different Luminance Value.”

This research explores how humans perceive noise intensity in digital color images and how luminance affects the ability to detect differences in noise. The study focuses on just-noticeable differences (JNDs) — the minimum perceptible variations that the human visual system can detect — under different levels of Gaussian noise and background brightness.

Psychophysical Study and Methodology

To analyze human perception, the team conducted a large-scale psychophysical experiment using a two-alternative forced-choice (2AFC) method. Participants were shown pairs of synthetic color image patches with controlled Gaussian noise and asked to select the noisier one.

Four luminance levels and four base noise intensities were tested, producing 16 reference conditions and more than 10,000 perceptual judgments collected across participants. The resulting data allowed the authors to derive psychometric functions and compute detection thresholds (JNDs) for each condition.

Key Findings and Insights

The study confirmed that noise visibility increases with noise intensity, consistent with Weber’s law — the higher the base noise, the greater the difference needed for detection. However, the relationship between luminance and JNDs proved more complex:

• At low noise levels, higher luminance masked the noise more effectively.
• At medium and high levels, the opposite occurred — noise became more noticeable around mid-luminance regions, suggesting adaptation effects in human vision.

These findings provide valuable insight into how noise masking operates across brightness levels and could inform more accurate visual difference predictors and image quality assessment models.

This work follows previous IMaLeVICS studies on noise perception and reduction, such as Image Noise Reduction by bootstrap resampling

The authors highlight the need for further experiments with noise levels closer to detection thresholds and for exploring non-uniform image regions (such as textures or edges). They also suggest testing alternative visual difference predictors to refine how psychophysical thresholds align with computational metrics.

This study contributes to a deeper understanding of how the human visual system perceives and adapts to image noise, bridging perceptual science and image processing. It also strengthens IMaLeVICS’ long-standing collaboration with institutions such as the University of Cambridge, University of Burgos, and University of Granada, advancing the group’s mission to integrate vision science and interpretable computational models.

Related approaches to image denoising using fuzzy logic were explored in Fuzzy Inference Systems to Fine-Tune a Local Eigenvector Image Smoothing Method