COGSCI 1 Lecture 9 - Perception Pt 1

The Turn to the Brain

early models of visualization focused on top-down analysis → in the 1980s scientists started studying the brain with the emergence of functional neuroimaging

Dorsal Stream

involved in the perception of spatial location; striate cortex → posterior parietal cortex

Ventral Stream

interconnected regions of visual cortex involved in
the perception of form; striate cortex → inferior temporal cortex

Marr’s Tri Level Hypothesis

emphasized the importance of understanding how cognitive processes are implemented in the brain;

Process for evaluating mental or artificial-information processing events

1. Computational Level

2. Algorithmic Level

3. Implementational Level

Computational Level

Highest, most abstract level

• What does the problem entail, i.e., what output is the system trying to get? What is the purpose or reason for the process?

Algorithmic Level

Programming Level

• Needs a formal procedure specifying how the data is to be transformed, what steps/step order

Implementational Level

Lowest Level

• Where is the hardware being used? How can the representations and algorithms be realized physically?

Marr’s Model of Visual Processing

built on a hierarchy of levels of studying cognition; System has to take a complex pattern of unstructured stimuli in the visual field and interpret them into representations that can then serve as input to more complex cognitive functions

First Stage (Marr’s Model of Visual Processing)

• Image is analyzed by intensity of light and dark areas

• Regions of sharp contrast indicate edges and contours → basic features of the object

• Raw primal sketch

Second Stage (Marr’s Model of Visual Processing)

• Features get grouped by similar size and orientation

• Processed to create a 2.5 D sketch

Third Stage (Marr’s Model of Visual Processing)

• Image is transformed into an object-centered 3D sketch

  • Symmetry axis: divides an object into mirror image halves

  • Elongation axis: defines direction along which main bulk/mass of a shape is distributed

Input from the Retina

conveyed via the optic nerve → optic chiasm → superior colliculus/brainstem → lateral geniculate nucleus (LGN) /thalamus

Area V1/Primary Visual Cortex

Where information processing begins; Neurons are sensitive to low-level features of the visual field (eg. orientation; direction of movement); LGN Projects here

Striate Cortex

V1 projects information here; anatomically distinct region of the brain

Area V2

Neurons have the same features but also some more complex ones (edges, shapes, depth); Area V1 projects here

Retinal Disparity

objects at different distances will fall on different locations on the two retinas; closer object = larger disparity

Extrastriate Cortex

region surrounding the striate cortex; processes additional features of visual information (movement, spatial frequency, retinal disparity, and color)

Trichromatic Theory

Retina has three color receptor types; each sensitive to red, green, and blue light

Opponent-Processing Theory

color-sensitive receptor cells respond in an opposing center-surround fashion to pairs of primary color

• Excited by light in the center; inhibited by light in the surround

• excited by light of a particular color; inhibited by light of the opposing color

For Ventral Pathway:

• Information goes V2 → V4 → inferior temporal cortex (ITC)

• ITC includes specialized areas for face recognition (fusiform face area) and identification of the human body/body parts (fusiform body area)

Similarities between Human Visual System and Neural Networks

• Information processing in the visual cortex is hierarchically organized — like neural networks

• Some parts are retinotopically organized — like convolutional neural networks

Blindsight

condition where people are blind in one or both visual fields —due to damaged visual cortex— but can “guess” (location or identity of objects; image details) accurately

Proposed Explanation for Blindsight

second visual pathway that does not go through the visual system; directly to the emotional/instinctual centers of the brain

How is Information Stored?

the majority of research supports the analog code (image), but some people on some tasks use propositional code (description)

Computer Vision Applications

Widespread use in areas like security surveillance, autonomous vehicles, and medical imaging

• Facial recognition systems in China; highly accurate; can recognize emotions

Ethical Concerns of Facial Recognition Systems

Found to be worse on women of color than white men

Ekmann Seven Basic Universal Emotions

Surprise, Anger, Fear, Contempt, Happiness, Disgust, Sadness

• Children who are born blind and/or deaf manifest emotions in the same way

Reading Emotions

Our brains are fairly accurate in reading emotions, we can evaluate people after 90 seconds; Emotions also conveyed through tone