Attentive texture similarity as a categorization task: Comparing texture synthesis models.

Area of Science:

• Computer Vision
• Human Perception
• Image Analysis

Background:

• Characterizing latent structures in texture spaces is challenging.
• Attentive similarity judgments are used to define perceptual texture spaces.
• An optimal description of perceptual texture space remains elusive.

Purpose of the Study:

• To establish a standard for relating image statistics to high-level similarity using human judgments.
• To evaluate how well different parametric texture synthesis models capture human perceptual similarity.
• To compare model performance against human data for natural textures.

Main Methods:

• Subjects grouped natural textures into visually similar clusters.
• Images were represented using features from three parametric texture synthesis models.
• Linear discriminant analysis predicted cluster membership based on model features.

Main Results:

• Model performance was compared against human cluster assignments.
• Comparisons were made for both positive and contrast-negated textures.
• The study evaluated the relative performance of different texture synthesis models.

Conclusions:

• Human similarity judgments provide a valuable benchmark for texture perception models.
• This approach aids in understanding the relationship between image statistics and perceived similarity.
• Further refinement of texture synthesis models can be guided by perceptual data.