200 likes | 449 Views
Semantic consistency versus perceptual salience in visual scenes: Findings from change detection. Spotorno , S., Tatler , B. W., & Faure, S. (2013). Acta Psychological, 142(2) , 168-176 . .
E N D
Semantic consistency versus perceptual salience in visual scenes: Findings from change detection Spotorno, S., Tatler, B. W., & Faure, S. (2013). ActaPsychological, 142(2), 168-176.
Objects are primarily selected on a perceptual basis with subsequent and supplementary effect of semantic consistency
Introduction • The allocation of attention during scene perception is likely to be guided by both low-level perceptual and higher-level information.
Semanticvs.Perceptual • Detection can be easier for changes that are visually salient (Pringle et al., 2001, Spotorno and Faure, 2011a and Spotorno and Faure, 2011b) • Contextual meaning and semantic relations within a scene can also play a substantial role when allocating attention toward the changes and in enhancing memory. (Stirk& Underwood, 2007, Hollingworth & Henderson, 2000)
Semantic • “central interest”——An advantage for detecting changes that involve the most consistent objects in the scene.( Auvray and O'Regan, 2003 , Kelley et al., 2003 , O'Regan et al., 2000 , Pringle et al., 2001 , Rensink et al., 1997 and Rensink et al., 2000 ) • semantically inconsistent objects or changes may be prioritised.( Gordon, 2004 , Gordon, 2006 , Hollingworth and Henderson, 2000 , Hollingworth and Henderson, 2003 and Stirk and Underwood, 2007 )
Perceptual • Some support for the notion that overt selection correlates with visual conspicuity, or visual salience ( Itti & Koch, 2000 ) has been found in scene viewing studies. (eg, Mananet al., 1997 , Parkhurst et al., 2002 and Reinagel and Zador, 1999 ) • Evidence for better or faster change detection for objects of high perceptual salience, either in conditions of overt ( Pringle et al., 2001 ) or covert ( Spotorno and Faure, 2011a and Spotorno and Faure, 2011b ) attention.
Materials • 16 coloured drawings, plus 4 for practice(PFLI — Bortolini, 1995 ) • 2 versions ——diagnostic or inconsistent • All the scenes were evaluated by 8 independent observers • 4 observers viewed the “diagnostic object” other 4 viewed the “inconsistent object” versions of the images.
Up to 3 objects ——diagnostic or inconsistent • Rate on a 6-point Likert scale the probability of occurrence in the scene (from 1, minimum, to 6, maximum) of a set of 7 objects pre-selected by the experimenters.
Half (8) of the diagnostic objects occurred among the first 3 most salient elements of the image. • The other 8 were not selected by the software among the first 7, and therefore we considered them as low salient objects.
One-Shot paradigm 1.. Trial examples. (1A) Deletion of a low salience/high consistency (i.e., diagnostic) item. (1B) Addition of a high salience/high inconsistency item.
Result • There was a main effect of Salience 80.32% vs. 49.7%
Result • no significant main effects for Type of Change or Consistency Diagnostic: 65.91%Inconsistent: 64.11 64.84% 65.17%
Result • Consistency and Salience interacted significantly • (High saliency)Performance was better for diagnostic than for inconsistent changes (82.96% vs. 77.69%)
Result • The interaction between Type of Change and Consistency was significant 68.73% 61.62% 63.09% 66.6%
Repetition of scenes and changes • The first changing version of the scene was detected with lower accuracy (56.74%) • Accuracy was no different between the second (64.94%), third (67.29%) or fourth (68.85%) change for each scene
main effect: Salience, Type of Change • Interaction: Consistency and SalienceF(1,30)= 4.23,p = .049
main effect: Salience, Type of Change • Interaction: Consistency and SalienceF(1,30)= 4.23,p = .049
This study was the first to compare directly objects that are diagnostic for the scene's gist and those that are inconsistent with it within a single experiment.
Semantic consistency in change detection • presentation times • types of images • Degree of clutter
Semantic analysis of perceptually salient objects reveals an inconsistency with the gist, further processing time is required. • a specific difficulty for the addition of highly salient inconsistent objects.