Depth variation and stereo processing tasks in natural scenes.

Area of Science:

• Computational vision
• Perceptual psychology
• Stereopsis

Background:

• Natural scenes exhibit local depth variations due to object edges and surface textures.
• The perceptual impact of these depth variations on stereopsis is not well understood.

Purpose of the Study:

• To investigate how natural depth variation affects performance in half-occlusion and disparity detection.
• To quantify local depth variation using disparity contrast in natural stereo images.

Main Methods:

• Utilized a large database of natural stereo images with laser-based depth measurements.
• Developed a method to sample stereo image patches and quantify local depth variation via disparity contrast.
• Analyzed the relationship between disparity contrast and performance in stereopsis tasks.

Main Results:

• Increased disparity contrast significantly degrades performance in half-occlusion and disparity detection.
• Disparity contrast alters the spatial integration areas (receptive fields) crucial for optimal performance.
• A simple binocular image statistic effectively predicts natural scene disparity contrast.

Conclusions:

• Local depth variation in natural scenes poses challenges for stereopsis.
• Findings provide insights into the mechanisms of stereo processing in natural environments.
• Motivates further computational and psychophysical research on local stereo perception.