Topic 9b: Objects, size, and space Flashcards
Summary of previous content
Luminance edges – V1,
Texture edges – V1,
Colour – V1, V2, V4 and V8,
Motion – V5/MT,
After-effects (inhibition over time) and contrast-effects (inhibition over space).
Big Question: What does this have to do with what we see?
What cortical pathways are involved in object recognition?
V1,V2 and V4 + top down: Segmentation.
V4: corners and proto-parts (SHAPE)
Posterior LOC (towards the back): parts of objects.
Anterior LOC (towards the front): whole objects and object recognition.
LOC - combining shapes into whole objects
What is the classic “bottom-up” view of vision?
- Processing starts with the stimulus on the retina and works via increasingly complex stages in ever higher brain areas towards some behaviour goal.
- For object recognition: first segment them from the background, then identify contours, then build contours into object outlines, then recognise objects.
Figure-ground segmentation
- Which neurons are responsible for this and where?
- Is this 100% bottom-up?
Segmentation can be based on texture, motion, depth cues provided by different types of RGC with different receptive field properties
Contrast sensitive neurons in each sub system:
- Retinal ganglion / LGN cells find object edges,
- Orientation-contrast -> texture segmentation,
- Motion-contrast -> figure-ground for motion,
- Disparity neurons -> direct object segmentation
You need top-down to assign figure and ground
How do illusory contours show top-down influence?
When does this develop?
Which cortical area is active and how?
- Perceiving a contour where no line exists. Top-down influences fill in missing information
- Develops at 8 month of age for static displays, 2-3 months if moving
- Contour sensitive cells in V1/V2 are active
Hyper-complex cells in V1 on either side of the illusory contour are inhibited by the extended lines, whereas those at the ends of the real lines are less inhibited. Receptive fields at line ending will response more strongly = perception of contour.
Thus, V1 cells detect the line endings and pass this signal to cells in V2.
What is contour integration?
How does it work?
Which area is active?
When does it develop?
- Ability to link up short sections of a contour to form a longer contour (helps us find objects).
- Opposite of length sensitivity in hyper-complex cells.
- Thought to come about due to long-range excitatory connections between cells that prefer similar orientations, but at different positions.
- Area V1.
- Develops around 2-1/2 yrs.
What are ‘form’ cells in area V4?
In terms of the processing of form. Neurons in area V4 respond to more complex features than earlier visual areas, but at a level less complex than whole objects.
Note, that contours can be of any type: real, illusory, defined by disparity (i.e., depth), motion or texture.
There seems to be a big role for attention in V4 (i.e., top-down influence).
How do we get from form to whole objects? Biederman’s recognition-by-components theory
Visual input is matched against structural representations of objects in the brain (i.e., geons).
Geons (geometric atoms) are the simple 2D or 3D forms such as cylinders, bricks, wedges, cones, circles and rectangles corresponding to simple parts of objects.
36 different geons for vision (cf 44 phonemes for speech).
LOC and object identification
LOC is certainly involved in shape recognition, but does it also identify specific objects (my cup, your watch)?
LOC is big:
Occupies a good part of the inferior temporal lobe.
Contains many sub-areas.
Close to face processing areas.
The necessary features of objects are all available to LOC.
So yes LOC could be doing specific object identification.
Some sub-areas seem to be responsible for object ID.
But, working out exactly what LOC does and how it does it is one of the current big challenges.
The problem is compounded by the fact human and monkey brains diverge at V4.
Top-down view
It’s a lot easier to find objects if you already know what it is (i.e., recognition precedes segmentation).
There is now evidence that human vision may also do this (e.g., it is possible to classify objects into categories very quickly if you know what you are looking for).
In this alternative view early visual areas provide a quick gist of an image to higher areas.
These higher areas then ‘guess’ what objects may be present and pass templates back down to the lower areas.
The lower areas then segment the objects and refine the locations of object contours.
The higher areas then get more information to make a better guess.
So, we know the gist of what something is before we know where it is.
The action of attention on V4 may be the point where this top-down influence is felt.
What does Shepard’s table top illusion say about processing?
The two table tops look like they are completely different shapes.
In fact, they are identical 2D shapes; at different orientations.
But, the brain produces different 3D shapes for the two tables, because the 2D line drawings of the tables are NOT identical.
Thus, the table tops are REQUIRED to be perceived as different shapes.
The ‘stubborn refusal’ of the visual system to see the two table tops as the same shape highlights something very important: we do NOT ‘see’ what’s in the retinal image.
OBJECTS Summary
Early visual areas segment the image, but possibly with the aid to top down influences based on an initial guess of object identify.
Illusory contours detected in V2.
All types of contour come together in V4.
V4 detects parts of objects like corners.
LOC processes whole objects, but is a large, complex, and not well understood region.
LOC probably also does object identification.
Where does place recognition occur?
- where stuff is in relation to you
- where you are now
Where stuff is in relation to you - dorsal “where” area
Where you are now - Ventral “what” area
Place recognition occurs in Parahipocampal Place Area (PPA)
What is size constancy?
Head size differs on the retina for people in the front row compared with those in the back row, but we know people’s heads are the same size
Beuchet Chair + Ames Room: Size Constancy Defeated
What is forced perspective?
Size constancy illusions (Beuchet Chair + Ames Room) make thing look closer than they really are, so we do not scale up the retinal image (i.e., perceive distance is too small, so we perceive things smaller than they are).
Forced perspective is a technique which employs optical illusion to make an object appear farther away, closer, larger or smaller than it actually is.
