Overview
This lecture explores how the human visual system constructs our perception of reality, examining the hierarchical organization from the eye to the brain, comparing biological vision to artificial intelligence (AI) models, and emphasizing the role of prior knowledge and top-down processing.
Visual Perception and Reality
- Our perception of reality is constructed by the brain, filling in gaps to create seamless experiences.
- Visual perception does not perfectly mirror external reality; the brain compensates for missing information, such as the blind spot.
Anatomy and Hierarchy of the Visual System
- Light enters the eye, is focused by the lens onto the retina, and detected by photoreceptors (rods for light, cones for color).
- The fovea provides sharp central vision; the periphery is less detailed and more sensitive to light.
- The optic nerve transmits signals from the retina, crossing at the optic chiasm, to the thalamus (lateral geniculate nucleus), then to the primary visual cortex (V1).
- Visual processing is organized hierarchically: simple features are processed in V1, progressing to complex forms and recognition in higher areas.
Functional Specialization and Streams
- The visual cortex is organized into dorsal (motion/spatial) and ventral (object/face/color) streams.
- Early visual cortex (V1) is retinotopic, overemphasizing foveal input, and processes basic features.
- Higher areas (V2, V4, V5) process increasingly complex aspects such as color (V4), motion (V5), and faces/objects (ventral stream).
Feature Selectivity and Complex Processing
- Simple cells in V1 respond to bars of light at specific orientations.
- Complex cells integrate input from multiple simple cells to detect motion/direction.
- Specialized areas like FFA (fusiform face area) are selectively responsive to faces; PPA for places/scenes; LOC for objects; EBA for bodies.
Correlation vs. Causation in Brain Function
- Experiments can show correlation (e.g., FFA activity with faces) but causality is determined through direct manipulation (e.g., brain stimulation).
Multisensory Integration and Abstraction
- Higher visual areas integrate input from multiple senses (e.g., FFA active during both seeing and touching faces).
- Representations become more abstract up the ventral stream, from concrete features to identity regardless of viewpoint.
Comparisons with AI Vision Models
- Deep neural networks (DNNs) and the brain share a hierarchical complexity, with early layers/regions detecting simple features and higher ones integrating complexity.
- AI errors highlight the unique role of human prior knowledge and expectations in perception.
Top-Down Processing and Prior Knowledge
- Perception is shaped by expectations and context (top-down processing), altering interpretation of sensory input.
- The brain contains extensive feedback connections, unlike traditional feedforward neural networks.
Temporal Dynamics and Recurrent Processing
- Visual processing unfolds over time, with information looping between posterior and anterior brain regions.
- Recurrent neural networks (RNNs) in AI mimic temporal processing but lack the adaptive complexity of human cognition.
Key Terms & Definitions
- Retina — Layer at the back of the eye containing photoreceptors (rods and cones).
- Fovea — Central region of the retina with highest visual acuity.
- Optic Chiasm — Point where optic nerve fibers cross, splitting information to opposite hemispheres.
- Lateral Geniculate Nucleus (LGN) — Thalamic structure relaying visual signals to the cortex.
- Primary Visual Cortex (V1) — First cortical area for visual input processing.
- Retinotopy — Spatial mapping of visual input from retina to cortex.
- Dorsal Stream — Visual pathway for motion and spatial information.
- Ventral Stream — Visual pathway for object and face recognition.
- Fusiform Face Area (FFA) — Brain region selectively responsive to faces.
- Top-Down Processing — Influence of expectations and prior knowledge on perception.
- Deep Neural Network (DNN) — AI model with layered processing, analogous to visual hierarchy.
Action Items / Next Steps
- Review assigned papers on multisensory integration (see provided journal links).
- Reflect on additional examples of how context and expectation shape perception.
- Prepare for discussion on causality vs. correlation in brain research.