L5.2 faces

Overview of Face Recognition

• Face Recognition is a central topic in PSYC 236: Cognition and Perception.

• Core Citation: Kietzmann, T. C., Poltoratski, S., König, P., Blake, R., Tong, F., & Ling, S. (2015). The occipital face area is causally involved in facial viewpoint perception. Journal of Neuroscience, 35(50), 16398-16403.

• The study of face recognition involves evaluating biological and cognitive evidence to determine if faces are processed through unique "special" mechanisms or general object recognition processes.

Evolutionary and Cognitive Theories of Object Recognition

• Major cognitive theories serve as the foundation for broader visual perception:     - Marr's Computational Approach (1982): Vision is viewed as staged information processing, moving from a primal sketch to a 2.5D representation, and finally a 3D model.     - Biederman's Recognition-by-Components (1987): Objects are recognized via "geons" (geometric units), prioritizing viewpoint invariance.

• Neuroscientific Theories include:     - Ventral vs. Dorsal Pathways: Functional specialization known as "What vs. Where/How."     - Global vs. Local Processing: Grounded in Navon (1777).     - Predictive Coding / Bayesian Brain: The concept that perception is a process of inference rather than just passive reception.

Thomas Bayes (c.1701 – April 7, 1761)

• Biography:     - English statistician, philosopher, and Presbyterian minister.     - Son of London minister Joshua Bayes; likely born in Hertfordshire.     - Enrolled at the University of Edinburgh in 1719 to study logic and theology.     - Return to London around 1722 to assist his father; moved to Tunbridge Wells, Kent around 1734.     - Minister of Mount Sion Chapel until 1752.     - Elected as a Fellow of the Royal Society in 1742.

• Published Works (during lifetime):     - Divine Benevolence, or an Attempt to Prove That the Principal End of the Divine Providence and Government is the Happiness of His Creatures (1731).     - An Introduction to the Doctrine of Fluxions, and a Defence of the Mathematicians Against the Objections of the Author of The Analyst (1736); published anonymously to defend Isaac Newton's calculus against George Berkeley.

• Legacy: His most famous accomplishment, Bayes' Theorem, was edited and published posthumously by Richard Price.

Bayesian Perception and Predictive Coding

• Core Theory: The brain acts as a "prediction machine."

• Inference Components:     - Sensory Evidence (Likelihood): Data detected by the senses, which is often noisy.     - Priors: Knowledge derived from past experiences, context, and expectations.     - Outcome (Posterior): The final "best guess" of reality.

• Mathematical/Statistical Framework:     - Prediction Error: The difference between the prior expectation and the peak of the likelihood function (reality).     - Uncertainty: Represented as the variance ($\sigma^2$) of the prior distribution.     - Noise: Represented as the variance ($\sigma^2$) of the likelihood function.

• Social and Environmental Scenarios:     - Example (Messy Desk): Identifying a cup involves a high prior (usually keep a cup there), a likelihood based on visible roundness/handle, and a posterior result identifying it as a cup. If the object is a container, the prediction error increases.     - Ambiguity and Dominant Priors: When evidence is noisy (e.g., poor lighting, fast movement), priors dominate. In social perception, stereotypes or confirmation bias can lead to false memories, such as an eyewitness recalling a weapon even if none was present.

• Garlichs & Blank (2024) Study:     - Used fMRI analysis to show scene cues predicted specific faces.     - Facilitation Effect: Expected faces are recognized faster.     - Assimilation Effect: Ambiguous morphed faces are often judged to be the expected face.     - Neural findings: Reduced activation for expected faces in the FFA (Fusiform Face Area), suggesting expectations reduce neural effort. Prediction error processing dominates the hierarchy moving from Oxygenated Face Area (OFA) to FFA to Anterior Temporal Lobe (ATL).

Behavioural Markers: Is Face Recognition Special?

• Face recognition is characterized by Holistic Processing, which is the simultaneous integration of multiple features into a single perceptual representation.

• Key Experimental Tasks:     - Part-Whole Task (Tanaka & Farah, 1993): Recognition of a specific face part (e.g., nose) is significantly more accurate when presented within the context of the whole face rather than in isolation.     - Face Inversion Task (Yin, 1969): Upright faces are recognized significantly better than inverted ones; the performance drop for inverted faces is much larger than the drop observed for inverted objects (like houses).     - Standard Composite Task (Hole, 1994; Young et al., 1987): It is difficult to perceive only half a face (e.g., top half) if it is aligned with a different bottom half. This effect is reduced when faces are presented upside-down.

• Nuances of Holistic Processing:     - Evidence from Rezlescu et al. (2017, 2018) suggests these three effects do not always correlate across individuals and show dissociations in neuropsychological cases.     - The "Holistic Family" includes: Inversion effect (Configural sensitivity), Part-whole effect (Contextual integration), and Composite effect (Perceptual integration).

Neuroscientific Markers of Face Recognition

• N170: An electrophysiological component (ERP) that shows a distinct negative peak around $170\,ms$ after stimulus onset; the peak is significantly larger for faces than for objects like cars.

• Face-Selective Network (Duchaine & Yovel, 2015):     - OFA (Occipital Face Area): Early perception of facial features.     - FFA (Fusiform Face Area): Perception of unique identity (invariant aspects).     - pSTS-FA / aSTS-FA (Superior Temporal Sulcus Face Areas): Processing changeable aspects (eye gaze, expression, lip movement).     - IFG-FA (Inferior Frontal Gyrus Face Area)     - ATL-FA (Anterior Temporal Lobe Face Area): Personal identity, name, and biographical info.

• The Expertise Hypothesis vs. Modularity Hypothesis:     - Modularity: FFA is a dedicated module exclusively for faces.     - Expertise: FFA activation is driven by any category requiring fine-grained individuation where the observer has expertise.     - Expertise Evidence: Gauthier & Tarr (1997) used "Greebles" to show FFA activation following training. Specific expertise (car experts, bird experts, chess masters, and radiologists) also recruits face-selective areas.     - Burns et al. (2019) Meta-analysis: Rejects the idea that expertise effects are due to p-hacking; supports the idea that the FFA plays a broader role in expert-level recognition.

Prosopagnosia ("Face Blindness")

• Acquired Prosopagnosia: Caused by brain damage; specific areas like the FFA are often involved.

• Developmental Prosopagnosia: No obvious brain damage, but presents similarly. Jiahui et al. (2018) noted reduced face selectivity in FFA.

• Selective Impairment (Patient GG): Intact recognition for birds, boats, cars, and chairs, but severe impairment for faces; points to a face-specific rather than general impairment (Busigny et al., 2010).

• Double Dissociation: While many have impaired faces/intact objects, some cases show the reverse (intact faces/impaired objects). However, Geskin and Behrmann (2018) found $20\%$ of developmental prosopagnosics have intact object recognition, suggesting face recognition may simply be a more difficult task requiring finer discrimination.

Face Recognition Models

• Bruce and Young (1986):     - Components: Structural encoding, Expression analysis, Facial speech analysis, Directed visual processing, Face Recognition Units (FRUs - structural info about known faces), Person Identity Nodes (PINs - biographical info), Name generation, and the Cognitive system.     - Pro: Explains name-recall problems and dissociations between identity/emotion (e.g., Capgras syndrome).     - Con: Focuses on static faces; stage-based/linear; underestimates interaction.

• Interactive Activation and Competition (IAC) Model (Burton, Bruce, & Johnston, 1990):     - A parallel, interactive computational model involving NRUs (Name Recognition Units), FRUs, PINs, and SIUs (Semantic Information Units).     - Pro: Explains priming and contextual facilitation.

• Distributed Human Neural System Model (Haxby et al., 2000):     - Core System: OFA, FFA, pSTS-FA.     - Extended System: Amygdala/Insula (Emotion), Intraparietal sulcus (Attention), Auditory cortex (Speech).     - Evaluates the ventral stream (identity) vs. dorsal stream (expression).

• Face Space Model (Valentine, 1991):     - Faces are represented as points in a multi-dimensional space.     - Center: Represents a hypothetical average face; typical faces are clustered here.     - Periphery: Distinctive faces are further out.     - Applications: Explains own-race bias, caricatures, and the distinctiveness effect.