61 terms

Lecture 6: Object recognition


Terms in this set (...)

Simplified to two pathways
visual processing in dorsal/ ventral pathways. Dorsal is where an object is. ventral is what an object is. Once the visual info reaches V1, the primary visual cortex ,does low level orientations of objects and features, then that is analyzed for higher level recognition of what and where the object is
What vs Where experiment
Monkey experiment. Lesioned a brain region of a monkey. a food was hidden under an object shape. monkeys had to know what each object was to find the food under it. Monkeys had to know which of the two identical triangles is located closed to the pole. Monkey with ventral lesion did poorly in object discrimination task. Monkey with dorsal lesion did poorly for landmark discrimination
Double Dissociation (function and location)
Double dissociation allows you to determine the relationship between a function and a brain location.
Double dissociation in humans
Dr. finds patients already lesioned. Patient DF have a perceptual problem called visual form agnosia, whereas dorsal lesion patient RV has an action problem called optic ataxia. Patient DF has a ventral lesion and has trouble with the "what". Patient RV has a dorsal location and an issue with the "where".
Visual form Agnosia
ventral lesion, issue with determining what an object is
Optic Ataxia
Dorsal lesion, issue with determining where an object is
What vs Where experiment task two
behavioral task. for perception, the patient has to discriminate between two different objects which targets the ability to recognize what the object is. Task 2 action task was asked to pick up the object for allocation study. Patient DF failed Task 1 and vice versa
Object agnosia
difficulty recognizing objects. Patients with ventral lesions. Patient DF saw object on the screen and asked to recognize them she couldn't. When person had physical contact with objects and had tactile info, she could label them correctly. meaning it wasn't a naming problem
Patient DF with a ventral lesion
When asked to draw what was asked, she could. suggesting that the memory was intact. She couldn't copy other pictures or recognize her own objects after drawing it. couldn't recognize at visual object recognition.
Explicit matching task of Patient DF
Showed patient DF with a slot that would rotate. She had to rotate the card like the slot would rotate. she couldn't.
Action Task of Patient DF
Patient was asked to put the card in the slot. meaning that when she performed an action that made her match the orientation of the slot, she was able to do it. THIS MEANS SHE HAS PERCEPTION DEFICIT BUT NOT IN ACTION: VISUAL FORM AGNOSIA
Optic Ataxia in Patient RV
Dorsal lesion. difficulty with visually-guided reaching - make inappropriate grasping movements with hands. Patient RV could distinguish between the objects on the basis of perception but couldn't use that info to guide the grasping movement in the right way. could touch locations on her body, but when using vision as a guide, she fails.
Summary of patients studies
Suggest double dissociation in ventral/dorsal pathways and perception/ action tasks. Main point= Object recognition involves distinct pathways
Simplified two pathways
both pathways connected by a pair of longitudinal fasciculus. visual info goes from V1 to inferior temporal cortex through the inferior longitudinal fasciculus. visual info in the dorsal stream goes from V1 to posteroparietal cortex by way of the superior longitudinal fasciculus. These processes coming from the primary visual cortex to temporal or parietal cortexes are called feed forward visual processing.
Over simplification?
there are connections between them; feeds forward and back and all directions. Pathways have connection signals flow forward and backward. the two clear dissociation are influenced by differences in attention, task demands and other things.
Evidence for Dorsal/ventral pathways
1. Animal lesion studies, 2. Human Neuropsychological cases - e.g., patient DF & RV, 3. Single unit recordings, 4. Human neuroimaging studies with neurologically intact subjects
Single unit recording
Recordings from ventral vs Dorsal neurons. What= Temporal neurons fire for: color, shape, faces. Where= Parietal neurons fire for: Direction of motion, velocity, control of (movements) of attention To know that something is moving is to know that the same thing has changed location (where). neurons in the temporal lobe which correspond to the ventral stream and neurons in the parietal lobe which correspond to the dorsal stream have different stimulus they fire for. in the inferior temporal cortex they fire for the recognizable shapes of hand and specific objects like faces. show that it fires for what the object is. neurons in parietal cortex fire for speed of motion, movement of attention, and direction. related to knowing of a location of an object
Human neuroimaging study
remembering what vs where. Healthy normal adults= Position task-Are the objects in the same locations? Object task- Are these the same objects? Same stimuli, different task. Healthy adults have shown dissociation of what and where. Position task= they saw a display flash twice. had to respond if a location was the same regardless of their object. Object task= had to respond if the identity of objects were the same or different.
Human neuroimaging study part two
Used PET for the task. found that distinct regions of the brain were found when ppl used the position task and object task. inferior temporal regions are identification locations parietal regions are involved in position tasks
Human neuro-imaging study of face/location matching
patients had to match faces to an original picture .location matching task, participants had to match the location of a dot to the spontaneously presented third location of the dot. Position task activated dorsal. face matching task was found in the ventral area
color verses motion
Color= compare brain response of color vs. black and white squares=ventral pathway. Motion= compare brain response to stationary vs. moving black and white regions= dorsal pathway. ventral pathway involvement in color and dorsal pathway involved in motion.
Lateral occipital complex
Human neuroimaging study was done on the ventral lesion of patient DF which matched to LOC in normal patients. Lesion marked in blue. this location is similar to an area that responds to objects in fMRI scans of normal brains. this area is LOC= sensitive to object and object shape info. similarity of DR and normal adults in fMRI suggest that the ventral area is processing the what info in objects.
How do we recognize objects that are not all the same as one thing? what about at different viewpoints
object constancy, Grouping of features and parts of objects, Use of stored object knowledge
object constancy
We can recognize objects across changes in illumination, size, occlusion, and viewing position, even though they create very different images on retina. How do we keep the objects identity constantly in mind and recognize the object as the same object from various view points or sizes. Visual system can tell that this is the same car from a different viewpoint.
Computational problems in object recognition
These two bikes look so different, yet we know that they are both bikes. How do we recognize that these are still the same blue cars? Even though the car is in the shade we can recognize both images as he same car even though the illumination is different
Grouping of features and parts of objects
Gestalt principle. what are the principles that allow you recognize that the striped circle is its own object in front of a striped background. What makes us recognize that it is a continuous branch behind a log? The GESTALT Principle are the laws of grouping on how the brain perceives patterns.
Gestalt law of grouping
Proximity, similarity, closure, good continuation, good form. important for our visual system to determine what bits of an image belong together and distinguish different objects from each other.
close up we see stimulus elements that are near each other as part of one objects and things spread apart as different objects
things that resemble each other are parts of the same object. we see the left two as part of the same object but the other two as part of a different object
forms are enclosed by a border and ignore gaps in the border. this can happen with incomplete objects. tend to see pairs of lines that make a square, not something else
good continuation
when stimulus elements move in the same direction and the same rate, they are the same
Good Form
the perceptual system tries to produce prospects that are elegant, which are symmetrical, predictable and regular.
Use of stored object knowledge
We can still identify objects even when there aren't local cues for grouping. Top-down processing from stored object knowledge based on the physical image. called the bottom up process which starts from low level image details. when the local cues not available, we can use stored object knowledge: this is called top down processing to recognize image
Bottom up processing
Perception starts at the sensory input, the stimulus. Thus, perception can be described as data-driven. For example, there is a flower at the center of a person's field. The sight of the flower and all the information about the stimulus are carried from the retina to the visual cortex in the brain. The signal travels in one direction.
Top-down processing
How our brains make use of information that has already been brought into the brain by one or more of the sensory systems. Top-down processing is a cognitive process that initiates with our thoughts, which flow down to lower-level functions, such as the senses. If one gives a cue that triggers a stored object to help u see and image. if you don't see a dog but someone tells you "dog", u start to see it.
Top Down processing part 2
recognize the person in the picture but there are two images. one with a young lady looking up, or seeing an old lady looking down.
How do we achieve object recognition?
some theories in object recognition include= Template matching and recognition by components
Template matching theory
it hypothesizes that there is a stored template in an object. and when visual input comes in, it will match the input to the template. Compared input to stored template of an object. Then respond whether a match or not.
Problems in template matching
Problem is that when the visual input comes in that doesn't look like the template, you can still tell its a B. this theory can't account for the changes in the templates
Problems in template matching 2
the H and A are identical but we still see THE CAT. if the visual system is using a template, then the response to the input should be identical. but they are not cause we see two different ones from context
Problems in template matching 3
all examples of a chair but all look different. So how could their be a template for all of them.
Problems in template matching 4
cant account for variability of exemplars, new objects, changing viewpoints, the idea of a grandmother cell rejected. cant account how we can recognize variability of exemplars, new objects, how we see same object with different viewpoint. There may be cells that have a template for your grandma but it can be rejected when you think how costly it is to have so many different cells of things in your life. And it doesn't take into account your grandmas variable changes. Yet, common views can be recognized faster cause it matches your template, whereas other points of view may take longer cause you need to rotate your template.
Recognition by components
Suggests that objects are recognized by its components. It assumes that there are specific shapes that account for the alphabet of vision. these shapes are geons.
basic building blocks of all object recognition. Geons are based on 3D shapes that can be assembled and make many different objects. like alphabet of the vision. Objects= goons + spatial relationshipsDifferent kind of objects can be represented from changing the location of one shape. Just like letters can make different words with different meanings, geons can do the same. This theory is lest costly cause only need 36 geons. also, they are 3D and can account for different points of view.
Recognition by components limits
how do you differentiate individual dogs that all have the same geon? fine with general categories but not individuals.
Example of visual agnosia
Patient GS (stroke at age 30). could see visual details. Could not recognize objects. could not describe function of objects - could make appropriate hand gestures.
Patient GS video
He could see the detail of the image and guess the size right. he has memory of different functions and other objects but can't figure it out. even after given the answer, he can't recognize it as a lock. there is a separate process for visually recognizing an object that involves the ventral pathway and acting upon an object that involves dorsal pathway
Different types of Agnosia
apperceptive, integrative, associative, alexia, prosopagnosia
Apperceptive agnosia
Failure in basic perceptual processing. More posterior lesions: earlier in visual processing stream. Right-lateralized (people with right damage). researchers could make a categorization of different types of Agnosia perceptive Agnosia: a failure in basic perceptual processing. may perform normally for shape discrimination, but fails to see objects with limited stimulus info, like a line drawing with deleted images or it its shown from an unusual view point. A perceptive agnosia can be a problem in object constancy. the damage is on the posterior part and on the right hemisphere.
Integrative agnosia
- Unable to integrate features into parts,or parts of an object into a whole Damage in both Right/leI hemisphere. Have a problem in integrating features and parts of a whole object. a normal viewer would see 2 squares with a circle. patients with integrative agnosia, are not able to see the integrated object, they only draw parts and lines of each object. The patient would recognize each line, the characteristic shape of a dog. problem when some parts are invisible or missing.
Associative agnosia
can perceive objects with the visual system, but cannot understand or assign meaning to the objects. Left hemisphere damage; more anterior regions. Fail at Matching-by-function task. Can see visual details of an object but cant access naming or meaning of the object. Patients will choose the two left images of umbrellas because of similarity rather than meaning.
Category specific associative agnosia
Patient JBR. - Common Objects: 90% correct. Live things: 6% correct. Similar category-specific deficits reported in several additional patients. No problem recognizing common items but when presented with live things, he got things wrong.
What are the implications of category specific agnosia?
We interact more with inanimate things: possible additional action representation, perhaps in motor system? Recall patient G.S. motor activation to manmade objects. Left Premotor becomes active when viewing tools. we interact more with inanimate objects, so its possible that our motor system is engaged with seeing the object. this differs with animals. Also left cortex becomes more active when seeing things that require motor skills.
What are implications of category specific agnosia? 2
Visual areas are segregated by animate vs. inanimate objects. Clue 1: we've already seen some specificity in ventral visual stream. One theory of this is because areas involved in visual processing in animate vs. inanimate objects might be separated. shows that some areas of the brain activate when visualizing different things.
What are implications of category specific agnosia? clue 2
more recent studies have suggested large scale segregation in ventral stream for living vs. nonliving things. another support for animate vs inanimate objects. researchers have found that there is a large scale segregation in ventral stream for animate and inanimate objects. purple is animals, green is object. Because animate objects are more lateral, its is more likely to be a damage of the lateral area so more may forget animate objects
Summary of Category-specific agnosia
The two main explanations= - Motor system engaged when viewing inanimate objects, helping agnostics to identify these objects better than animate objects - Ventral visual system is segregated by animate vs. inanimate large-scale regions, as more recently shown by fMRI. - Based on vasculature, regions representing animate objects may be more prone to damage due to encephalitis or stroke (thus, the asymmetric loss of animate objects but not inanimate)
Summary in my words
The motor system is engaged more, there are two separate streams of processing in a ventral pathway that specialize in inanimate vs. animate stimuli. more patients have impairment in recognizing animals
"Face agnosia" or "Face blindness". Sometimes following damage to fusiform area. Sometimes congenial (from birth; runs in families). People know that there is a face there but don't know who the face belongs to. Can't link faces to memory. they can tell gender or age, hairstyle or clothing style but cant recognize face. This follows damage to the fusiform area. or it could be congenial, born as prosopagnosic. Can recognize that it is a face, but can not recognize the identity of a face. Uses non-facial info such as hair or clothes. This suggests that there is something special to faces, and nothing else.
Important study of faces
Subjects more accurate at identifying whole faces than parts. Accuracy is 60% when asked to see a part of a face, whereas it was much easier to see the face as a whole. this whole-effect wasn't true for other types of objects, just faces
Holistic processing of faces
Faces are perceived holistically and changes to parts of a face is hard to see when things are upside down.
Evidence from Agnosia suggesting face specificity
evidence for space specificity comes from a patient with object agnosia. patient CK cant recognize objects if shown figure on the right, he saw confusing colors and shapes. If the inverted figure was presented, he saw a face. His face recognition was spared compared to his object recognitions.