Object Perception CLPS 0020 Introduction to Cognitive Science
Object Perception CLPS 0020: Introduction to Cognitive Science Trent Wirth Fall 2016
Why study Perception? Remember Marr’s levels of analysis? • Perception is the input to the cognitive system – Computational Level: If you don’t understand the input, you’re a little lost
What is object recognition? • Our world is filled with complex scenes BB-8 R 2 -D 2
A more common example You do this ALL THE TIME
A simple model of object recognition Retinal image Object description Stored object representations Identification
Questions • What are the nature of object description and stored object representation? • How are the mappings carried out?
Early models of object recognition • Template matching • A bucket • Feature matching • Letter between G and I T E C T • Work for barcodes (by computers), but don’t exactly translate to human perception
Two Contemporary Models • Object-centered approach – Perceptual image is broken down and contrasted against a stored representation • Viewer-centered approach – Perceptual image is contrasted against an object representation that is dependent on the viewer’s orientation
Object-centered approach (1) • Marr et al (1978): Generalized Cylinders/Cones
General Approach / Problems • Object representations are built from these “building blocks” • Perceptual image builds up to generalized cones and compares against stored representations • Might be a good strategy to get computers to recognize objects – But little psychological evidence
Object-centered approach (2) • Biederman’s (1987, 2000) Geon theory (aka. “Recognition by Components”)
Geon Model • Set of basic shapes that were the building blocks of all object representations – Number of basic shapes changed over the years (initially 5, now 32 of them) • Object recognition is performed by breaking down visual representation into geons and comparing combinations with object representation – The more geons you could view, the easier object recognition would be
Supporting evidence for Geon model (Biederman et al, 1973)
Supporting evidence for Geon model (Biederman et al, 1973)
Thoughts on Geons • Geometric objects as building blocks – Intuitive! – more information = better recognition • Problems? – Very abstract – Where do these building blocks come from?
Contrast: Viewer-centered approach • Multiple views are stored for each object • One view is “Canonical” • Canonical views are faster to recognize • Transformational mechanism is required to compare perceptual image with object representation
Support for viewer-centered approach • Tarr (1995): Novel object learning • Two phases of experiment • Phase 1: Learn a novel object
Phase 2 • Classify new views of objects – Is this the same or different?
Predictions Difference in perspective
Tarr (1995) Data • Reaction time increased for categorizing object as the same in a linear fashion as difference
Thoughts • Studies of object recognition under multiple perspectives support the viewer-centered approach • Our brain probably has both kinds of representations – An Example: Faces (More on this in face recognition lecture)
Making sense from noise • What is it – Do you see the dog? • Once you see the dog, can you not see the dog? – Perceptual system makes sense of noise to form objects
Categorizing Sounds - Speech • Speech sounds can vary in a number of ways • One dimension along which consonants vary is Voice-Onset-Time (VOT) – /ba/ vs. /pa/ • Adults do not perceive all changes along this continuum equivalently – Hear categories of sounds
Categorical Speech Perception • Eimas (1973): Habituated Sucking Response • Play a particular sound categorized as /pa/ or /ba/ over and over again – Method: Sucking rate is equivalent to looking time. It increases then decreases – Then change the sound
Speech Categorization Data • Eimas et al. (1971): Distinction between /pa/ and /ba/ in 1 month-olds and newborns
Conclusions • Speech (phoneme) perception involves a great deal of noise – Different frequencies: pitches, loudnesses, amplitudes, but also levels of background noise, multiple sources, etc. • At birth (or at least by 1 mo), we are beginning to categorize those sounds into meaningful units – Basis of language learning?
Cue Combination • A final piece of the puzzle – https: //www. youtube. com/watch? v=Gl. N 8 v. Wm 3 m 0
- Slides: 27