Software Based Affective Smile Estimation
Translated title
Computer baseret affektiv smil genkendelse
Author
Bøttern, Tomas Fuchs
Term
4. term
Education
Publication year
2012
Submitted on
2012-05-24
Pages
108
Abstract
Dette speciale undersøger, om et computerprogram kan bedømme menneskers smil lige så godt som mennesker. Studiet angriber problemet fra computerens perspektiv og fokuserer på, hvad der kan udledes direkte af ansigtstræk i billeder. Vi gennemførte to tests, hvor programmet gav en smilvurdering af fotos, som vi sammenlignede med vurderinger fra testpersoner. I den første test havde programmet ret i cirka halvdelen af tilfældene (50%). I den anden test matchede programmets vurderinger den gennemsnitlige vurdering fra menneskegruppen. Programmet klarede sig godt, når smil havde tydelige og let genkendelige visuelle træk; det havde svært ved mere almindelige eller subtile smil og når betydningen især lå i blikket. Arbejdet trækker på affective computing (at lære computere at opfatte og reagere på menneskelige følelser), Facial Action Coding System (FACS) og dets aktionsenheder (betegnelser for specifikke ansigtsmuskelbevægelser) samt begrebet emotionel intelligens (at opfatte og ræsonnere om følelser). Samlet set kunne computeren kun vurdere smil pålideligt, når de var visuelt tydelige; når blikket dominerede, kunne den ikke vurdere smilet korrekt.
This thesis explores whether a computer program can judge human smiles as well as people do. The study approached the problem from the computer’s perspective, focusing on what could be extracted directly from facial features in images. We ran two tests in which the program assigned smile ratings to photos and we compared these to ratings from human participants. In the first test, the program was correct about half the time (50%). In the second test, its ratings matched the average ratings from the human group. The program performed well when smiles had clear, easily detectable visual cues; it struggled with more common or subtle smiles and when the meaning depended mainly on gaze (eye direction). The work draws on affective computing (teaching computers to sense and respond to human emotions), the Facial Action Coding System (FACS) and its action units (labels for specific facial muscle movements), and the idea of emotional intelligence (perceiving and reasoning about emotions). Overall, the computer could reliably rate only visually distinct smiles; when gaze dominated, it could not rate the smile appropriately.
[This abstract was generated with the help of AI]
Keywords
Documents