Face Recognition Lecture Notes - 3
Face Recognition
- Specialized topic following object recognition.
- Learning Objectives:
- Understand neural processes for face recognition.
- Identify dedicated brain regions for facial information.
- Understand the psychological basis of caricatures.
- Recognize limited accuracy in recognizing unfamiliar faces and its implications for eyewitness testimony.
Neuropsychology of Face Recognition
- Prosopagnosia: Failure to recognize faces.
- Differentiated from failure to recognize objects.
- Brain regions:
- Process faces.
- Process sight of bodies.
- Grandmother Cells: Theoretical concept of individual cells responding only to specific people (e.g., one's grandmother).
- Neural processing of faces is biased.
- Eyewitness testimony accuracy is limited.
Disorders of Face Recognition
- Prevalence:
- Developmental prosopagnosia affects approximately 3% of people.
- Can also result from brain damage.
- Catastrophic Recognition Loss: Inability to recognize oneself, family members.
- Brain Lesions:
- Typically disrupt outputs from the visual cortex to the temporal lobe.
Symptoms and Interpretations of Face Agnosia
- Symptoms:
- Failure to recognize individual faces.
- Recognition of others possible through cues like clothes, voice, or gait.
- Visual system in prosopagnosia:
- Face signals are not processed correctly.
- Interpretations:
- Specialized brain system for face recognition exists.
- Damage to this system compromises face recognition.
- Controversy exists regarding recognition of other object classes.
Challenges in Face Recognition
- Faces share common components.
- Differentiation relies on feature configuration.
- Patients with prosopagnosia may struggle to differentiate objects within a class (e.g., car models).
- Examples:
- Mechanic failing to recognize car models after brain lesion.
- Birder losing ability to recognize bird species.
Evidence for Specialized Face Recognition
- Patient WJ:
- Severe face recognition impairment (50% accuracy).
- Could recognize sheep from his flock.
- Study by McNeil and Warrington:
- Politician-turned-sheep farmer with impaired face recognition but intact sheep recognition.
- Farmers as a control group:
- Showed normal face recognition abilities.
Double Dissociation
- Patient C.K. (reported by Moskovitz):
- Could recognize faces but not the component vegetables in a face made of vegetables.
- Double Dissociation:** Demonstration of preserved ability in one domain and impaired ability in another, and vice versa in another patient.
- Evidence for separate brain mechanisms for faces and objects.
Brain Imaging and Face Recognition
- Functional Magnetic Resonance Imaging (fMRI):
- Identifies active brain areas.
- Nancy Kanwisher's Study:
- Scanned individuals viewing faces and objects.
- Subtracted brain activation to objects from activation to faces.
- FFA responds more to faces than objects, houses, or scrambled faces.
- Located in the right hemisphere on the fusiform gyrus in the temporal lobe.
- Active during comparisons between profile faces and hands.
- Small region (fingernail-sized) present in most individuals.
- Properties: Responds more to faces than other object classes, even those requiring expertise.
Extra Striate Body Area (EBA)
- Discovered by Paul Dunning in Cardiff.
- Responds to the sight of bodies, body parts (arms, feet, hands), people, line drawings, stick figures, silhouettes, and recumbent people.
- Does not respond to:
- Tools.
- Jumbled stick figures or silhouettes.
- Whole faces (intermediate responses to parts of the face).
- Bodies of other species (intermediate response).
Brain Regions and Specialization
- Orange region: Extra striate body region.
- Purple region: Fusiform face area.
- Parahippocampal place region (in the hippocampus): Responds to buildings and aids navigation.
- Different brain systems are specialized to respond to different stimuli.
Neural Processing Hierarchy
- First visual area:
- Simple, complex, and hypercomplex cells.
- Temporal cortex:
- Cells responding to elaborate shapes.
- Suggests a hierarchy of processing.
Selfridge's Pandemonium Model
- 1970s model describing potential brain function.
- Demons:
- Feature demons: Detect component lines of letters.
- Cognitive demons: Recognize combinations of lines.
- Decision demon: Identifies the letter based on strongest activation.
- Strictly hierarchical.
- Cognitive demons feed on to decision demon.
- More complex as you move up the hierarchy.
Grandmother Cells Revisited
- Hypothetical cells responding only to a specific individual (e.g., one's grandmother).
- Problem: Perceived lack of sufficient brain cells.
- Refutation: Cubic millimeter of brain contains approximately one million cells.
- Recognition vocabulary is limited (e.g., 20,000 words).
- Specificity makes finding these cells difficult.
Historical Developments in Monkey Brain Studies
- Areas analogous to the human fusiform face area exist in monkey brains.
- Doris Tsao's Recordings:
- Used micro electrodes to record from cells in this region.
- Tested selectivity of cells with different stimuli:
- Faces, fruits, bodies, gadgets, hands, and scrambles.
Selectivity of Cells in Monkey Brains
- 97% of cells in the recorded region were selectively responsive to faces.
- Correlation between fMRI studies and single-cell recordings:
- Active cells corresponded to active brain regions in fMRI.
- Back of the brain feature detectors.
- Posterior temporal cortex elaborate cells.
Anterior Temporal Cortex
- Cells respond to the sight of faces.
- Construction of faces from selectively picking of particular features using elaborate shapes.
- Cells generalize across:
- Position
- Size
- Orientation
- Lighting
Responses to Various Faces
- Cells respond to various faces.
- Sensitivity to perspective view:
- Respond to front views but not back views.
- Some cells respond only to profiles.
Criticism of Marr's and Biederman's Models
- Models are incorrect in their description of how the brain recognizes faces.
- Two-dimensional images analyzed from one perspective.
Selectivity for Individual Faces
- Cells respond more to specific individual faces under different viewing conditions.
Recordings from Epilepsy Patients
- Tom Otter's lectures discussed recordings from epilepsy patients.
- Neurosurgeons placed electrodes to find the origin of seizures and found cells responding to faces.
- Example: Cell responding to Jennifer Aniston's face.
Halle Berry Cell
- Cell responded to various images of Halle Berry, including her role as Catwoman with a mask.
- Responded to the name "Halle Berry" when written out.
- Concept of a person accessed through name, face, and body.
- Cells like this exist in the temporal cortex.
Sensitivity to Identity
- Small number of cells are sensitive to identity.
- Other cells are selective for:
- Familiar objects (e.g., food items).
- Famous buildings (e.g., Sydney Opera House).
- An array of cells responds to highly specific items an individual is familiar with.
Visual Cues for Face Recognition
- Facial features (eyes, nose, mouth).
- Eyebrows are a crucial cue.
- Eyes are less important than expected.
Internal vs. External Features
- Internal features.
- External features (hair) are important, especially for unfamiliar faces.
- Configuration of features is important.
Importance of Internal vs. External Features
- Unfamiliar faces: External features are crucial.
- Familiar faces: Internal features are more important.
Thatcher Effect
- Inverted faces with locally upright features appear normal until upright.
- Features are processed with respect to gravity.
- Recognition system is locked to gravity.
- Features are analyzed independently.
Caricatures
- Faces are marked with shape of the face and chin.
- Average face shape is constructed from 50 female faces.
- Original images are warped to align with the average face and create a grand average.
- Represents distinctive features of the class of stimuli.
Construction of Facial Averages
- Averages are made of white female adults and male politicians.
- Resultant image contains all of those traits.
- Can be recognized as a woman/white woman or a man.
- Individual instances of politicians used.
Exaggerating Differences from Average
- Faces are warped away from average.
- Elongated faces become more elongated.
- Exaggeration produces caricatures.
Caricatures and Recognition
- Images presented with names in a recognition task.
- Real photographs (no distortion): 1.35 seconds recognition time.
- Caricatures: More efficient recognition.
- Caricatures exaggerate deviations from average.
- Help recognition system improve its speed of recognition.
Coding Differences from Average
- Faces are coded individuals in the brain by codifying differences from average.
- Differences from average form the basis of recognition when meeting people.
Biases in Face Recognition
- Similarity Judgments:
- When asked to choose which face is more similar, the left one or the right one.
- Tend to pick based off of the majority because the right isn't like the other.
- Splitting middle picture in half.
- People use information on the left half of the face for similarity. Right half is not used.
- A mirror that halves had the double left looks more like the target than the double right.
- Face processing is biased.
Masculinity Judgments
- Task: Decide which image looks more masculine.
- Result: Most people pick the top image.
- Split Stimuli:
- Same image, one flipped over.
- Blend between male and female face.
- Conclusion:
- Attention is paid to features on the left side of the face.
Hemispheric Specialization
- Visual information from the left side projects to the right brain.
- Fusiform face area is in the right brain.
- Left side of the world has preferential access to the fusiform face area.
- Pay more attention to the left side of faces, which projects more to the right hemisphere.
Implications of Biased Face Processing
- Pay more attention to the left side of faces for:
- Age judgments
- Attractiveness judgments
- Left half of face projects to right hemisphere.
- Right hemisphere has specialized face processing equipment.
Challenges in Recognizing Unfamiliar Faces
- Difficulty matching passport photographs to real people.
- Matching Accuracy:
- Security and identifying the images to the siblings.
- Passport fraud.
- Checking ID.
Limitations of Eyewitness Testimony
- Eyewitnesses are not very accurate.
- High error rates (30%).
- No improvement with training.
- Police officers perform no better than students in matching photos to CCTV images.
- Super-recognizers exist, but normal training doesn't help.
Line-Up Identification
- Task: Identify the person from an array of faces.
- Problem: False positives can occur, even when the person is not present in the line-up.
- Eyewitness testimony is not always accurate. People can get falsely accused.
Key Takeaways
- Recognizing familiar individuals is excellent.
- Recognizing unfamiliar faces is poor.
- Beware eyewitness claims and false positives.
Further Exploration: Capgras Delusion
- Neuropsychology literature is full of interesting case histories.
- Capgras Delusion:
- A delusion where individuals believe that familiar people have been replaced by imposters.
- How do the symptoms of someone with Capgras resemble horror movies?
- Invasion of the Body Snatchers has similarities to Capgras delusions.