Low Level Visual Saliency Does Not Predict Change

Low-level visual Low-level visual saliency does not saliency does not predict changepredict change

detection in natural detection in natural scenesscenes

Stirk, A.,& Underwood, G.(2007). Journal of Vision, 7(10), 1-10.

• change blindness– Maintenance failed (Rensink,2002; Simons &

Levin, 1997)• Coherence field dissolves

– Rensink (2000)

• flicker– (Rensink ,1997)– A–blank–A’–blank-A

• Top–down and/or Bottom–up →allocation of attention

• Wright (2005)– change detection in natural scenes could be

predicted– subjective measures

• Influenced by top–down processes.

• When semantic information is low, bottom–up processes may have a greater influence on the allocation of attention

MethodsMethods

• 2x2 design• Salience (high level vs. low level)• Scene-schema(consistent vs. inconsiste

nt)

MethodsMethods

MethodsMethods

• 24 participants• 10scenes (19.7° × 13.9 °)

– 4 changed images– 1 original image

– 80 trials ：4*10 change pairs 、 4*10no-change pairs

ProcedureProcedure

• 按鍵回答 “ SAME” or “DIFFERENT”• 重複” Flicker” ，直到受試者做出反應

• 練習： 8 trials ( 有 Feedback)• 正式： 80 trials ( 沒有 Feedback)

ResultsResults

• Consistency RT ：– F(1, 23) = 5.38, p = .03– IC 2341.7 < C 2549.2

• Visual Saliency RT ：– F(1, 23) = 1.78, p = .20,– No main effect

• Consistency ACC ：– F(1, 23) = 15.55, p = .001– IC 87.3% > C 78.3%,

• Visual Saliency ACC ：– F(1, 23) = 0.26, p = .62,– No main effect

DiscussionDiscussion

• Inconsistent-object detection advantage

• Categories of objects guide visual attention

• Violations to the scene-schema→stronger perceptual

representation

• change detection based solely on the visual properties of a scene and finds that semantic salience.