Week 5 PDF - Visual Object Recognition

WEEK 5 Template theory- highly detailed description that must be matched perfectly for perception issues of size, orientation, and non standard presentation (biological plausibility). When we recognize an object we recognize it as a whole. ​We recognize objects by comparing them to stored "template...

WEEK 5 Template theory- highly detailed description that must be matched perfectly for perception issues of size, orientation, and non standard presentation (biological plausibility). When we recognize an object we recognize it as a whole. ​We recognize objects by comparing them to stored "templates" or mental representations in our memory. When we see an object, our brain tries to match it to one of these templates, much like a puzzle piece fitting into a specific shape. Each template represents a different object or a variation of an object. For example, if you see a letter "A" in a specific font, you have a mental template for that "A." When you see a different font or handwritten "A," your brain will still try to match it to your stored "A" templates. Consistency is important for this theory, requiring every object a person has ever seen in every orientation you've seen it in. The reason we know how different fonts of letters look is because our mind has templates of different fonts. (captcha, bots have a hard time recognizing words). (angle we see object in does matter) Feature analysis theory- when you recognize a chair ect, really what you're doing is visually breaking down the object into the parts that make it up and recognizing the object by parts. Rather than recognizing an object as a whole, this theory identifies a combination of features or perceptual recognition. Visually breaking down the object and then reassembling it. (angle we see a object in doesn't matter) Recognition by components RBC (biederman, 1987)- objects consist of 3D pieces that he calls “geons.” He thinks every object is made out of geometric shapes that he called geons. In your mind you have a dictionary of geons that are different from each other. To recognize an object, you have to think of how the geons come together. Insensitive to object orientation but sensitive to geon visibility. Doesn't care what angle the object is at as long as you can see the geons (lego pieces that make up the object), orientation doesn't matter. Sometimes, features other than shapes and geons are important to separate one object from another. Beidarman was focused on the physical aspect of shape. -one of the earliest things studied in brain was visual object recognition Repetition suppression (RS)- discovered shortly after fmri was developed in early 90’s. When you see a coffee cup our brain will activate and activate attention, visual areas. If you take the coffee cup away and show it again then your brain is less active the second time you recognize the object. More you see the exact same visual stimulation the less brain activity you have. Can use subtraction method, neural activity for presentation one subtracted by neural activity presentation 2. Our brain is fine tuning the features we are paying attention to and responding to but filtering things that don't matter (if colors don't matter for an umbrella, we're not gonna focus on the color areas on the brain but shape size etc matter). -N = Novel R = Repeated F = Fixation (nothing on screen -We also have RS when we show someone something in a different orientation or angle. If we show someone A coke bottle twice they have RS but if we show the same bottle in a different angle then the brain sees that New bottle as a new object (what template theory suggests). Angle and view matters and its called view Dependent processing, no RS and brain acts as its 2 completely different objects. -when there's RS it means you recognize both are coke bottles no matter what angle it's at. View-Dependent Theory View-dependent theory argues that our ability to recognize objects depends on seeing them from familiar viewpoints. If the object is viewed from a new or unusual angle, it might be harder to identify. View-Invariant Theory View-invariant theory suggests that we recognize objects regardless of the angle or viewpoint. So, even if you see an object from a different perspective, you can still identify it. -in the left fusiform area (part of the temporal lobe in the left hemisphere) pattern of findings we see. In this Area we get RS when we see the object at a different view (view does not matter view invariant) -right hemisphere, right fusiform area part of the temporal lobe known for object recognition. Same object Different views give no RS (view does matter, view dependant). After visual information reaches the occipital lobe (the part of the brain responsible for initial visual processing), it travels in two main pathways or "streams": 1. Ventral Stream (or "What Pathway"): This pathway moves from the occipital lobe toward the temporal lobe. It's called the "What Pathway" because it helps us identify what we are seeing—like recognizing a face, an object, or a word. This pathway is essential for object recognition because the temporal lobe, where it leads, is also involved in memory. We use memory to compare new information with what we already know, helping us recognize and identify things. -The where/how pathway is for motor controls which runs dorsal/superiorly from the occipital lobe to the Parietal cortex. Helps us construct 3 dimensional map of space so you know where objects are and how far Something is. Tell the motor system how to move to grab something. -people were interested in how we analyze objects, starting in the mid 1990’s. Nonsense objects are things That looks like objects but aren't actually objects. They have geometric form and complexity but you can't Recognize it as an object you've seen before. Compared brain activity when looking at objects like these To visual noise. Looking at parts of the brain that analyze form, temporal lobe. Professor martin would use The subtraction method with nonsense objects vs actual objects that make sense. He found even more Activity in the ventral stream, left and right hemisphere as well. When we analyze a form of the object or Identify it, we engage the ventral stream. -identifying faces, animals ect is from ventral stream -a valley activated ventral and dorsal stream. When we see a mountain, river we use ventral because its Visual aspect. For the spacial properties such as house, valley we start to analyze it using the dorsal stream Fusiform gyrus/ face area (FFA)- part of the ventral/ temporal lobe mostly for face recognition. Would only Lightup for faces and not light up for objects. Right side of the brain was more active (for no reason). People Who have autism often avoid eye contact and avoid looking at faces, their FFA may work differently. -The parahippocampal place area for spatial scenes -gattier wanted to show that the FFA does more than just face. They create play dough like creatures called Greebles that could be categorized as different people. Once you practiced with these greebles to where they Could recognize them. People who had never seen greebles before, when they see humans they’d activate The FFA. but when they see greebles they don't activate the FFA. couple weeks after interacting with the Greebles and becoming an expert you activate the FFA and use it for greebles. -car experts activate FFA for cars and faces both but not birds -bird experts will activate FFA for birds and faces but not cars We cannot tms (because it is at the bottom of the brain) FFA to confirm if it's actually responsible for what it does. When someone has a stroke in the posterior cerebral artery which covers occipital and bottom of Temporal lobe, they make damage the FFA which may cause prosopagnosia (inability to distinguish between faces/face blindness). Mostly happens with brain injury but also possible through birth (developmental prosopagnosia) happens in childhood. Visual agnosia- difficulty in identifying an object using vision. After damage to the ventral stream a person May be able to write but not recognize their own writing. Sees objects well but cannot use information to Identify, perception and memory is fine but cannot connect the 2 (disconnection of ventral stream). Have Hard time linking vision to object knowledge. When a person also impairs their perceptual abilities, we call it Apperceptive visual agnosia is caused by brain injury in posterior areas and is rare. They say it's worse than Being blind, can distinguish between light and dark but not more than that. -being born with or without vision and touching a object, does not change how knowledge is organized in the Ventral stream. Multi Voxel brain analysis (MVPA)- brain acts differently when seeing one thing vs another. is a brain imaging method that helps researchers understand how different patterns of brain activity relate to specific thoughts, perceptions, or tasks. When you think about something or do an activity, specific areas in your brain light up in unique patterns. MVPA takes the data from many small brain areas (called voxels) and analyzes them together to find patterns. This way, instead of looking at one small spot in the brain, researchers look at how multiple spots (or "multi-voxels") work together. By studying these patterns, scientists can see which brain patterns relate to certain activities, like recognizing faces or reading words. This helps them learn how different parts of the brain contribute to thoughts and actions. -you can scan someone's brain when they're looking at 100 houses and then you get a template for what your brain looks like when it's looking at a house, then you put it through a software that knows how to decode brain activity for houses. Same goes with faces. If you show that software a face or a house, it will predict what you're looking at based on what you taught it (68 percent sure you're looking at a house right now). You can use this to know what someone is thinking of or dreaming of based on these patterns.