Object Recognition Flashcards
Perception is more than information, what else does it involve?
Recognition and interpretation
Perceptual Constancy
The idea that we perceive constant properties of the world.
Unconscious inference
The idea that a stimulus can be perceived in more than one way
Likelihood principle
The idea that when we perceive an object we do so in a manner that is most likely to have caused the pattern based on our experiences with similar situations in the past.
Gestalt Laws of Organization
- Law of Good Continuation
- Law of Simplicity
- Law of Similarity
- Law of Familiarity
Law of Good Continuation
Law of Simplicity
Law of Similarity
Law of Familiarity
True or False
Our brain first collects information and then interprets it.
False
Our brain collects information and interprets it in parallel.
Brain areas for simple features and large scale configuration work together and interact to achieve recognition.
What do we mean when we say that Gestalt Laws are heuristic?
It means that Gestalt Laws are most of the time right but could be wrong.
Physical Regularities
- Vertical and Horizontal lines are better recognised than oblique lines
- Light comes from above
Semantic Regularities
The meaning
Bottom-Up influences
Information that we get from the stimulus itself
Top-Down influences
Information the perceiver brings to perception.
Word Superiority Effect
Words are much easier to recognise than individual letters.
This tells us that it is easier to recognize things when they are in the correct context.
True or False
If you are shown a nonsense word that slightly looks like an English word, you would be able to recognize it.
True
This is because of pronounceability work as context.
It could also be because of the probabilities of letter combination.
For example, TAS, RAF are more probable combinations than QWR, TKM
Feature Net
A feature net is how you would break something into smaller components such that you can recognise it. For example, the word CLOCK you can break it into possible combinations of letters, then into individual letters and each individual letter can be broken into further component such as line segments with different orientations.
Detectors
In our brain, we have multiple detectors that are responsible for detecting a different feature of the object to recognise. For example. if we want to recognize the word CLOCK we would first have detectors picking up the features composing each letter, then detectors picking up each letter and finally a word detector.
How does a detector is activated?
Sufficiently strong signal or an accumulation of many smaller signals that will hit a response threshold and thus make the detector fire.
True or False
Detectors are the same a neurons
False
Detectors do not refer to neurons
What determines the starting activation level of detectors?
- recency: If a detector has fired recently
- familiarity: How frequently a detector has been used
True or False
The bigram level in the feature net allows us to be more efficient when recognizing a combination of letters and will always be right
False
The bigram level in the feature net allows us to be more efficient when recognizing a combination of letters but it makes our recognition less accurate.
What our brain actually does is to read a few letters and then make some inferences on what the word could be.
Recognition by Components (RBC) model
This is another model that can help us understand object recognition. In this model, we have geons or geometric ions which are basic geometric shapes that can compose any object.
True or False
We only need three dozens of geons to compose any object in the world
True!
This was discovered by Irving Biedeman
Is there a problem with the RCB model?
Yes, it is point view dependent
Fusiform brain area
Area of the brain dedicated to processing faces and other very familiar objects.
Mirror Neurons
Neurons that fire when we see someone performing an action as if we would be performing the action too.
