Textures vs Non-Textures: A Simple Computational Method for Classifying Perceived 'Texturality' in Natural Images

Area of Science:

• Visual Perception
• Computational Neuroscience
• Image Processing

Background:

• The human visual system efficiently processes scenes and surfaces using statistical representations of image textures.
• The definition and objective identification of 'textures' in natural images have remained largely subjective.
• Understanding the empirical basis of texture perception is crucial for visual system research.

Purpose of the Study:

• To empirically investigate the conditions under which natural images are perceived as textures.
• To establish an objective, quantitative link between image statistics and perceived texturality.
• To develop a reliable method for determining if an image region is processed as a texture by the visual system.

Main Methods:

• Correlation analysis between perceived texturality and image similarity to synthesized images using the Portilla-Simoncelli (PS) model.
• Statistical analysis of image properties, focusing on specific PS statistics.
• Development and validation of a discriminant model using image statistics to classify perceived textures.

Main Results:

• Perceived texturality strongly correlates with the similarity between original images and their PS synthesized counterparts.
• Both perceived texturality and similarity judgments are highly correlated with specific PS image statistics.
• A discriminant model achieved over 90% accuracy in discriminating images perceived as textures.

Conclusions:

• Image texturality can be objectively quantified and is linked to specific statistical image properties.
• The Portilla-Simoncelli (PS) model and its associated statistics provide a basis for understanding texture perception.
• This research offers a method to determine if image regions are statistically represented by the human visual system.