Perception of objects in natural scenes and the role of attention (Part 2) Karla Evans & Anne Treisman Princeton UniversityAddressing two questions Test feature priming hypothesis as a possible explanation for rapid scene categorization Test the attention capacity available for visual categorization in natural scenesParadigm Trial duration: 450 and 720 ms in exp.1&2 1320 ms in exp.3-6Testable predictions Performance should deteriorate when the non-target scenes share some of the same features with targets. Uncertainty about the identity of the detected target. Detected targets could often be wrongly located. Inversion of the scene will leave intact the interference from people distractors.Animal Targets Vehicle Targets 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Non-Human Distractors Human Distractors Non-Human Distractors Human Distractors Conditions Prediction 1 (Experiment 1) 75ms image exposureDetected: 73% Animals Identified: 53% (e.g. as Ferrari, orfreight train) Classified: 78% ( e.g. as mammal,or bird) Located: 53% (left, right or center) Detected: 74% Vehicles Of those detected: Identified: 43% (e.g. as bear, orsnake) Classified: 84% ( e.g. as car, orplane) Located: 56% (left, right or center) Prediction 2 & 3 (Experiment 1)Upright Images Inverted Images 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Non-Human Distractors Human Distractors Non-Human Distractors Human Distractors Conditions Prediction 4 (Experiment 2) 110ms image exposureRole of attention in natural scene categorization (Exp.3-6) T1 and T2 (blocked). mixed). identify T2. Experiment 3- AB classical design, identify Experiment 4- identify T1 and T2 (randomly Experiment 5- only detect T1. Report and Experiment 6- only detect both T1 and T2.Experiment 3 (identify T1 and T2 -blocked) 40% 50% 60% 70% 80% 90% 100% 220 660 220 660 Percentages CorrectVehicle Animal 440 880 440 880 T1 to T2 Lag in ms Single Dual Same Dual DifferentCategory known vs. unknown a) Category known (Exp.3) b) Category unknown (Exp.4) 40% 50% 60% 70% 80% 90% 100% 220 440 660 880 220 440 660 880 T1 to T2 Lag in ms Dual Same Dual DifferentIdentifying versus Detecting T1 Identify T1 (Exp. 3) Detect T1 (Exp. 5) 40% 50% 60% 70% 80% 90% 100% 110% 220 440 660 880 220 440 660 880 Lag in ms Dual Same Dual DifferentIdentifying versus Detecting T2 Detect T2 (Exp.7) Identify T2 (Exp.5) 40% 50% 60% 70% 80% 90% 100% 220 440 660 880 220 440 660 880 Lags in ms Dual Same Dual Different (Exp. 6) (Exp. 4)Summary Early aspects of natural scene categorization may reflect the parallel detection of disjunctive sets of features rather than the binding and individuation of high-level objects (exp.1& 2) Identification of a category target requires attention and competes with detection of a second target appearing within the next 800-1000ms. (exp.3-6)Thank