DOC PREVIEW
UNCW PSY 211 - Pattern Recognition Cont'd

This preview shows page 1 out of 4 pages.

Save
View full document
View full document
Premium Document
Do you want full access? Go Premium and unlock all 4 pages.
Access to all documents
Download any document
Ad free experience
Premium Document
Do you want full access? Go Premium and unlock all 4 pages.
Access to all documents
Download any document
Ad free experience

Unformatted text preview:

PSY 211 1st Edition Lecture 10 Outline of Current Lecture I. Local (Feature-based) vs. Global (Holistic) Processing.II. Cognitive Theories of Pattern Recognition III. The Cognitive Neuroscience of Pattern RecognitionCurrent LectureLocal (Feature-based) vs. Global (Holistic) ProcessingLeft Hemisphere- Local Processing- Processing Features- “The trees”o You see individual trees before you see the forest.Right Hemisphere- Global Processing- Processing of holistic patterns.- “The forest”o You see the forest before you see individual trees.Is Pattern Recognition is based on Feature or Holistic Representation?Evidence that PR is Feature-driven1. Features-detection cells in primary visual cortex2. “Pop-Out” effects in Visual SearchEvidence that PR is driven by Holistic Representation1. Top-down influences on PR (expectancy (“set”) effects in perception; semantic priming).2. Object-detection (“grandmother”) cells in inferior-temporal (IT) cortex (far along the “what” pathway).3. Face inversion effects.These notes represent a detailed interpretation of the professor’s lecture. GradeBuddy is best used as a supplement to your own notes, not as a substitute.Evidence for Feature-Based PR:Feature Detection Cells- Cells in early visual areas (V1-V4) respond selectively to basic ("primitive") features, including line orientations, colors, angles, sizes, intensities, spatial frequencies, and movements.Pop-out Effect in Visual SearchCognitive research shows that if a search target is defined in terms of a single primitive feature, it "pops out" allowing many items to be searched at once (known as "parallel search"). In contrast, if a search target is defined by a conjunction of primitive features, then each item mustbe searched one at a time (known as "serial [or conjunction] search").Evidence for Holistic-based PR:Object-detection ("grandmother") cells in inferior-temporal cortex. These are cells that represent complex but specific concept or objects. It activates when a person “sees, hears, or otherwise sensibly discriminates” a specific entity. Face Inversion effects- we have developed a way of looking at face holistically. Cognitive Theories of Pattern RecognitionTemplate Matching ApproachesInput image is maintained as holistic pattern and match to “templates” (holistic reps) in memory. It is widely used in machine-recognition systems. Template Matching as limitations such as:- It requires exact math between image and template- Inputs very widely in size, orientation, etc.- Inputs are often degraded or occluded- For real-world application, a large number of templates would be required.Feature Extraction ApproachesPandemonium"Feature demons" become active when their feature is presented, thereby activating "cognitive demons". A "decision demon" decides what letter was presented. Limitations of Feature-detector Models like Pandemonium1. Completely bottom-up (does not allow priming)a. No chance for context/expectation or higher-level representation to influence thePR of low level features. 2. Relations among features are losta. Not easy to distinguish between patterns that share multiple features. (e.g., O vs.Q, E vs. F, F vs. P, etc.).Feature Network ModelsJust like neurons, feature detectors have a baseline activation (resting) level and a response (firing) threshold. Some detectors fire more easily than others due to frequent or recent input (cf. 'priming'). A strong input or several weak inputs can cause a detector to fire. If a detector fires, it activates detectors at the next level in the network. Feature network models have a number of interesting properties that are consistent with phenomena observed in human PR including (a) the tradeoff between accuracy and efficiency; (b) the effects of "wellformedness" on PR; and (c) the way we know much more than we've been taught (based on the idea that knowledge is "distributed" -- embedded in the connection strengths between detectors).Interactive ApproachesConnectionist Model (Parallel distributed Processing (PDP) Models)Units: neuron-like processing nodes that take on values.Connections: the links between units (cf. axons); connections have weights (+/-) that determineshow a unit is affected by activationHidden Units: units that have no connection with the outside world (cf. interneurons).Key advantages of PDP networks is their ability to learn & generalize, to complete partial/messy patterns, and to show "graceful degradation" – the ability to still function when damaged (whenunits are removed/injured).The McClelland and Rumelhart PDP Model- Bottom-up and top-down connections- Features and (in some models) holistic reps.- Excitatory and inhibitory links- Activation (and thus recognition) occurs gradually over time, allowing higher-level (e.g., word) units to "guide" lower-level (e.g., letter & letter feature) units. -Recognition by ComponentsTheory of 3D Object Recognition that proposes the existence of object primitives (“geons”) which combine to form objects.Interactive models (PDP and Geons) get around the limitation of the other models by:- Allowing both bottom-up & top-down processes to operate quickly and simultaneously.- Allowing holistic representations ("templates") to influence/facilitate/guide the processing of features.The Cognitive Neuroscience of Pattern


View Full Document

UNCW PSY 211 - Pattern Recognition Cont'd

Download Pattern Recognition Cont'd
Our administrator received your request to download this document. We will send you the file to your email shortly.
Loading Unlocking...
Login

Join to view Pattern Recognition Cont'd and access 3M+ class-specific study document.

or
We will never post anything without your permission.
Don't have an account?
Sign Up

Join to view Pattern Recognition Cont'd 2 2 and access 3M+ class-specific study document.

or

By creating an account you agree to our Privacy Policy and Terms Of Use

Already a member?