High level perception - Object perception

Question 1

what are the principles of the gestalt psychologists?

Answer 1

“The whole is more than the sum of the parts” —> There are grouping principles of perceptual organisation

Question 2

what are the five grouping principles?

Answer 2

similarity
proximity
closure
good continuation
common fate

Question 3

closure

Answer 3

Area bounded by closure is seen as a separate object - closing and area of space

Question 4

good continuation

Answer 4

Good Continuation - continuity of lines to see them together

Question 5

common fate

Answer 5

when they move in the same direction and belonging together

Question 6

Figure ground

Answer 6

Figure ground - border creates two separate objects - like the vase with two faces

Question 7

what type of model is Marrs model of recognition, what does it aim to explain ?

Answer 7

This is a computation model - processing and different stages

Attempts to explain how we recognise objects in the visual field.

Question 8

what are the stages of Marrs model of recognition?

Answer 8

primal sketch
21/2 D sketch
3-d Model

Question 9

primal sketch

Answer 9

What happens: Marr’s model begins with the detection of edges in the visual scene. These edges represent changes in intensity (light and dark) and are key for identifying boundaries of objects.

Why it’s important: Edges are the first step in understanding the basic structure of the objects in our environment. These edges are processed using edge filters, which detect contours and shape.

Question 10

2.5 D

Answer 10

What happens: After detecting edges, the brain groups them into larger structures or objects. This is where Gestalt principles (such as proximity, similarity, and continuity) come into play, as the visual system organizes the edges into coherent shapes and surfaces. It begins to form a 2.5D sketch, which includes the shape, depth, and orientation of the object relative to the observer.

Why it’s important: This step is about organizing and simplifying the visual input into meaningful chunks that help us perceive objects in three-dimensional space.

Question 11

3D model

Answer 11

What happens: In this stage, the visual information is combined and abstracted to create a more detailed, 3D representation of the object. The model is not specific to any one viewpoint but provides a flexible structure that can be recognized from different perspectives.

Why it’s important: This 3D representation allows for recognition from any angle or orientation, creating a more robust model for object recognition.

Question 12

how do we recognise an object according to Marrs model?

Answer 12

What happens: Finally, the model of the object is compared to stored memory models in long-term memory (such as from previous experiences or learned knowledge). This comparison allows us to recognize and categorize the object.

Why it’s important: We match the newly processed object model with existing models stored in memory, which helps us quickly identify and understand the object.

Question 13

principle axis

Answer 13

biggest cylinder in shape

Question 14

limitation of Marrs

Answer 14

Problem - many objects are hard to recognise upside down

Question 15

strengths of Marrs

Answer 15

Marr’s model predicts that:

Visibility of the principal axis is crucial for recognition. If it’s hidden, recognition is harder.

Once a 3D model is built using the axis, different orientations of the object are equally easy to recognize (viewpoint-invariant).

🔍 This explains why we struggle to recognize objects when their core structure is obscured, and why we can still recognize familiar objects from unusual angles.

Question 16

Biederman Recognition by components

Answer 16

1. Edge extraction
   Surface characteristics: luminance, texture, colour
2. Detect arrangement of edges:
   Curvature, parallel, co-terminating, symmetry, co-linear,
   Such arrangements do not alter with view
3. Segment object into components (parts)
   Detect concave parts
4. Determine GEON type for each component (part)
   36 GEONS needed

Question 17

Geon

Answer 17

Geon is a shape defined by whether the axis is straight or curved- the cross section (angular or smooth or symmetric)

Question 18

strength of Biederman

Answer 18

Model properties - Need only few geons to describe most objects (3 geons = 150 million possible objects) Recognition is viewpoint invariant

Question 19

limitations of Biederman

Answer 19

Problems - Does not differentiate objects within class (how do you tell the difference between your mug and your friends mug) , Does not use surface pattern, Recognition is viewpoint invariant - evidence suggests otherwise - view sometimes matter on how you see an object (Tarr & Bulthoff, 1995)

Question 20

visual processing in v1 goes where

Answer 20

goes in two directions - temporal lobe and parietal lobe

Question 21

when monkeys had a lesion to the temporal lobe did they struggle with pointing out the new object (one they had not been shown before) or did they struggle with reaching the food near the object?

Answer 21

They cannot work out which object is the new object

Question 22

when monkeys had a lesion in the parietal lobe what did they struggle with?

Answer 22

They couldn’t reach for the food near the object

Question 23

where is damage in the brain for people with object agnosia ?

Answer 23

temporal lobe

Question 24

what are the symptoms of object agnosia?

Answer 24

No loss of intelligence
Failure to recognize objects
No simple visual impairment
Can see edges but cannot put them together
May draw object OK but not recognize drawing

Question 25

Patient DF

Answer 25

who suffered carbon monoxide poisoning - tested for object/size/shape and orientation - she couldn’t describe them but she could pick them up and guide her hand to the object. could not indicate size of an object

Question 26

Milner’s post task

Answer 26

Line up the plate with slot - she couldn’t match the angle but she could just post it in - she cannot describe orientation

Question 27

Titchener circles - illusion effect

Answer 27

2 circles surrounded by different sized circles we think the the circle surrounded by smaller circles is bigger than the one surrounded by larger circlers. Even though physically the same.

Question 28

when a device measured people's hands reaching for a circle (one surrounded by small circles and one surrounded by bigger circles) what did they find?

Answer 28

the hand size was unaffected by the illusion even if you thought that one of the circles was smaller the width of your hand to grab this would be the same as to grab the other circle. This suggests that —> Object analysis for action is separate from conscious perception of object

Question 29

optic ataxia symptoms

Answer 29

- Difficulty completing visually guided reaching tasks - Difficulty reaching in the right direction - Difficulty positioning fingers correctly towards an object - Little relationship between grip aperture and object size

Question 30

what stream

Answer 30

Ventral - Temporal lobe

Question 31

where stream

Answer 31

Dorsal - Parietal lobe

Question 32

what do early models of vision assume?

Answer 32

👁️ Vision helps the brain build a picture of the world inside your head. 🧠 This helps you: Know what things are Know where things are Think and act based on what you see ✅ The goal of vision is to understand and move around the world.Early models assume purpose of vision is to construct and internal model of reality - foundation for all visually derived thought and action

Question 33

what brain area allows us to recognise the same object in more than one location?

Answer 33

Lateral Occipital Cortex (LOC) = identity representation

Question 34

Hubel and Wiesels findings in visual cortex

Answer 34

simple —> complex —> hypercomplex

Question 35

what is the bottom up flow of visual info in the brain ?

Answer 35

V1 (edges) (simple) - V2 (contours) - V4 ( colour and shape) - PIT (simple features) - AIT ( elaborate features) down the temporal lobe becoming more complex

Question 36

Simple vs. Complex Cells

Answer 36

Simple Cells (V1): Detect edges, orientations Complex Cells (V2): Detect contours, motion Hypercomplex Cells (V4+): Detect colour, shape, complex objects

Question 37

Posterior vs. Anterior Temporal Cortex

Answer 37

Posterior: Sensitive to orientation & size Anterior: Less sensitive to size/orientation, focuses on object identity

Question 38

Object Representation in the Brain

Answer 38

Cells generalise across position Organised in columns Complexity increases down the temporal lobe

Question 39

Single-Cell Representation of Objects

Answer 39

Individual neurons respond to specific object features. Cells generalise across position but increase in complexity down the temporal lobe. Early areas (V1) detect edges, later areas (AIT) recognise whole objects.

Question 40

Max operation

Answer 40

which one are you responding the most - no matter of location more and more selective of things in the environment Helps provide invariance (ensures the system still recognizes an object even if it changes position, size, or orientation).

Question 41

Bottom-up processing

Answer 41

single cell representations Starts with simple features (edges in V1). Progresses to complex features (shapes & objects in AIT). Info flows from basic to detailed, building full object recognition.

Question 42

Top-down processing

Answer 42

Memorised concepts - high level object detectors- mid level pattern = context

Question 43

does more info come down or go up the system?

Answer 43

10 x comes down the system - more descend from temporal cortex

Question 44

Object permanence

Answer 44

We expect an object moved behind a screen to reappear with the same form when screen removed. less than 6 month olds will be shocked that the object has gone.

Question 45

object representation

Answer 45

‘Similarity’ underlies how objects are represented, Similarity guides classification, naming, behaviour, What makes objects similar or dissimilar?, Two objects similar = neural representation similar

Question 46

the importance of context for object recognition?

Answer 46

Recognition of objects is easier in correct context

Question 47

Word superiority effect

Answer 47

Detecting a letter in is easier when in a word

Question 48

V1: Simple & Complex Cells

Answer 48

Detect edges & orientations. MAX operation: Provides invariance to small changes in position.

Question 49

V2: Composite Features

Answer 49

Combines V1 outputs to detect shapes & contours. Helps in pattern selectivity.

Question 50

PIT: View-Tuned Cells

Answer 50

Recognises whole objects (e.g., airplanes, faces). Uses a weighted sum (template matching) for better recognition.

Question 51

Hierarchical Processing in Object Recognition - the word C

Answer 51

Low-level: Detects simple features (edges, curves). Mid-level: Combines features into patterns (letters). High-level: Recognizes whole objects/words (“CAT”). 🔼 Neurons fire stronger as patterns become more meaningful!

Question 52

IS OBJECT RECOGNITION JUST BUTTOM UP OR TOP DOWN?

Answer 52

its bidirectional

Question 53

ventral visual cortex activity

Answer 53

The ventral visual cortex is responsible for recognizing and categorizing objects, and its activity is influenced by how similar objects are perceived to be.

Question 54

in brain study what representations are first to emerge?

Answer 54

object colour shape background

Question 55

marrs object identifcation summary

Answer 55

Edge filters analyze visual information (detecting edges). Gestalt grouping organizes these edges into meaningful shapes. The object is then represented as a cylinder-like model (3D shape). The model is compared to stored memory representations of objects to recognize and categorize it.

Question 56

problem of occlusion

Answer 56

the challenge of understanding and perceiving objects that are partially or completely hidden by other objects in the environment.

Question 57

What does PIT stand for in visual neuroscience?

Answer 57

Posterior Inferotemporal Cortex

Question 58

What is the primary function of the PIT?

Answer 58

Object recognition in the ventral visual stream (the "what" pathway).

Question 59

What does "weighted sum process" mean in the context of PIT?

Answer 59

It refers to the integration of inputs from lower visual areas, where each input has a different importance or weight.

Question 60

Why are inputs "weighted" in the PIT?

Answer 60

To give more influence to certain features that are more relevant for recognizing complex objects.

Question 61

What does the PIT achieve by using a weighted sum of inputs?

Answer 61

Increased pattern selectivity, allowing it to recognize objects despite minor variations (like size, rotation, or partial occlusion).

Question 62

What type of information does PIT pool together?

Answer 62

Feature information from lower-level visual areas like V1, V2, and V4.

Question 63

How does the weighted sum process help in object constancy?

Answer 63

By combining features in a selective way, the brain can recognize the same object even if it looks slightly different.

Question 64

What is the role of the MAX operation in V1 complex cells?

Answer 64

It selects the strongest (maximum) response from multiple simple cells detecting the same orientation at different positions.

Question 65

Why is the MAX operation important in visual processing?

Answer 65

It provides position invariance, meaning the brain can still recognize a feature (like an edge) even if it moves slightly.

Question 66

How do complex cells in V1 use the MAX operation?

Answer 66

They pool responses from several simple cells and take the maximum one to determine if a specific edge orientation is present, regardless of exact location.

Question 67

What does "distributed object representations" in the ventral visual cortex mean?

Answer 67

It means that object features are processed across different regions of the ventral stream. Each region handles specific aspects (like color, shape, texture), and these features combine to form a complete mental image of the object.

Question 68

How does perceived similarity affect object recognition in the ventral visual cortex?

Answer 68

Objects that are perceived as similar (e.g., two animals) activate overlapping regions in the ventral visual cortex, helping the brain categorize them based on shared features like shape, texture, and color.

Question 69

What is the significance of similarity-based object representations in the brain?

Answer 69

The brain groups objects with similar features into overlapping patterns in the ventral cortex, facilitating recognition and categorization. Objects with similar characteristics activate common brain regions, aiding in efficient object processing.