{[ promptMessage ]}

Bookmark it

{[ promptMessage ]}

Object Recognition S08p - EXP 3604 Classifying Concepts...

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: EXP 3604 Classifying Concepts Classifying One of these things is not like the other … or is it? Object Recognition • Def: The identification of a complex Def: arrangement of sensory stimuli arrangement • Leads to identification of a stimulus as a Leads whole, meaningful object or event whole, • Vast number of distinct patterns can be Vast learned learned • Can be very fast and “automatic” • 3 categories: patterns, objects, faces Theories of object recognition • • • Feature Analysis theories RBC theory View-Based approaches – Template View-Based Theory Theory Feature Analysis Theories • We store critical features that differentiate stimuli Demo: Find the Letter M Biederman's RBC theory • We recognize objects by – decomposing them into geons – specifying relations among geons specifying relations GEONS OBJECTS Biederman's RBC theory • Biederman & Blickle Biederman (1985) (1985) – Recognition of fragmented Recognition objects: intersections vs. continuous edges continuous Results: Results – Recognition depends on Recognition ease of activating geons ease Template theories • Stimuli matched to a set of templates (patterns) stored in memory = MATCH BUT… = NO MATCH = NO MATCH = NO MATCH Context and Object Recognition • Bottom-up vs. Top-down Processing – Examples: Word Superiority Effect • People are better at recognizing letters in People the context of words than letters presented by themselves. presented W O R K vs. _ _ _ K vs. O R W K • Why? WORD Top-down processing LETTER Bottom-up processing FEATURE SENTENCE Top-down processing WORD Bottom-up processing LETTER Face Recognition • Tanaka & Farah (1993) – Learned names that were paired with faces or houses – Tested on recognition with whole or parts. Tanaka & Farah (1993) • Results: Percent correct 90 Isolated-part condition Whole-object condition 80 70 60 50 Faces Houses • Conclude: – Faces are processed holistically. Tanaka & Farah (1993) • Another experiment used upright vs. inverted faces during learning. – Results: Percent correct 90 Isolated-part condition Whole-object condition 80 70 60 50 Upright Faces Inverted Faces – Conclude: Inverted faces disrupt holistic processing. ...
View Full Document

{[ snackBarMessage ]}