Abstract
The efficient coding hypothesis posits that sensory systems are tuned to the regularities of their natural input. The statistics of natural image databases have been the topic of many studies, which have revealed biases in the distribution of orientations that are related to neural representations as well as behavior in psychophysical tasks. However, commonly used natural image databases contain images taken with a camera with a planar image sensor and limited field of view. Thus, these images do not incorporate the physical properties of the visual system and its active use reflecting body and eye movements. Here, we investigate quantitatively, whether the active use of the visual system influences image statistics across the visual field by simulating visual behaviors in an avatar in a naturalistic virtual environment. Images with a field of view of 120° were generated during exploration of a virtual forest environment both for a human and cat avatar. The physical properties of the visual system were taken into account by projecting the images onto idealized retinas according to models of the eyes' geometrical optics. Crucially, different active gaze behaviors were simulated to obtain image ensembles that allow investigating the consequences of active visual behaviors on the statistics of the input to the visual system. In the central visual field, the statistics of the virtual images matched photographic images regarding their power spectra and a bias in edge orientations toward cardinal directions. At larger eccentricities, the cardinal bias was superimposed with a gradually increasing radial bias. The strength of this effect depends on the active visual behavior and the physical properties of the eye. There were also significant differences between the upper and lower visual field, which became stronger depending on how the environment was actively sampled. Taken together, the results show that quantitatively relating natural image statistics to neural representations and psychophysical behavior requires not only to take the structure of the environment into account, but also the physical properties of the visual system, and its active use in behavior.
Introduction: The Efficient Coding Hypothesis
The efficient coding hypothesis suggests that our brains are optimized to process the specific types of visual information we encounter most often. To study this, researchers typically analyze databases of "natural images." However, traditional databases often rely on planar sensors and limited fields of view, failing to account for the physical properties of biological visual systems and the active nature of vision—including body and eye movements.
Methodology: Virtual Forest and Active Vision
To address these gaps, researchers utilized the Unity 3D engine to construct a photorealistic virtual forest. This environment allowed for the simulation of "active" observers with species-specific heights and gaze behaviors.
The study employed two avatars: a human (1.8m height) and a cat (0.25m height). Three gaze strategies were simulated:
- Straight: Looking directly ahead in the direction of movement.
- Down: Looking 30° downward (simulating navigation over rough terrain).
- Random: Gaze directions sampled from a normal distribution.
Crucially, the 3D environment was projected onto idealized spherical retinas using geometrical optics models for both humans and cats. This approach allowed for the analysis of image patches across a 120° field of view (FOV), examining power spectra and edge orientation histograms to identify statistical biases.
Key Findings: Geometry and Behavior
The study yielded several critical insights into how anatomy and behavior shape visual input:
Validation of the Virtual Environment
The virtual forest successfully replicated the second-order statistics of the real world. In central vision, the VR data matched the 1/f² power law and the "cardinal bias" (predominance of horizontal and vertical edges) found in established photographic databases like the van Hateren dataset.
The Emergence of Radial Bias
A significant finding was the emergence of a "radial bias" in the periphery. As eccentricity increases, orientations tend to align with lines radiating from the center of the gaze. This effect is a geometric consequence of projecting 3D space onto a sensor and is more pronounced at larger eccentricities (e.g., 50° vs. 30°).
Behavioral Influence on Statistics
Active gaze strategies radically altered the statistical signatures. The "Down" gaze strategy, common during navigation, shifted the dominant orientations from the vertical lines of trees to the complex textures of ground cover and perspective lines converging at a vanishing point.
Species-Specific Input
Anatomy also played a role. Due to their proximity to the ground, the cat avatar's lower visual field contained significantly higher "power" (detail and contrast) compared to the human avatar, simply because ground-level objects were physically closer to the eye.
Conclusion
The research demonstrates that natural image statistics are not solely a property of the environment but are deeply intertwined with the observer's physical properties and behavior. Understanding neural coding requires a move toward "active" observer models that account for retinal geometry and behavioral gaze strategies. Future work should extend these findings to diverse environments and incorporate more complex biological factors, such as varying photoreceptor density and optical aberrations.
Summary updated