A scale invariant measure of clutter

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Perceptual Psychology

Background:

• Visual search performance is influenced by image complexity.
• Quantifying image clutter is crucial for predicting search efficiency.
• Previous clutter measures often lack scale-invariance.

Purpose of the Study:

• To propose a novel, scale-invariant measure of image clutter.
• To evaluate the predictive power of this clutter measure on visual search times.
• To establish a quantitative link between image properties and search behavior.

Main Methods:

• Utilized an efficient image segmentation algorithm to generate multi-scale segmentations.
• Modeled the relationship between region count and segmentation scale using a power law.
• Determined image clutter by fitting power law functions and using the constant of proportionality.
• Conducted a visual search experiment using 160 real images as stimuli.

Main Results:

• A scale-invariant measure of image clutter was successfully derived.
• The proposed clutter measure accounted for approximately 40% of the variance in visual search times.
• The power law exponent relating region count and scale was consistent across images.

Conclusions:

• The developed image clutter measure offers a robust predictor of visual search performance.
• Scale-invariant clutter quantification can improve models of visual attention and search.
• This approach provides a valuable tool for optimizing image display and search interfaces.