Feature Level Processing Lessons from low-level vision Applications in Highlighting Icon (symbol) design Glyph design Spotfire product Visual symbols Architecture for visual thinking Primitives of Perception (the phonemes). The whole visual field is processed in parallel This machinery tells us what kinds of information are easily distinguished Popout effects (general attention) Segmentation effects (dividing up the visual field) Segmentation by Primitive Features stereo form color form stereo form color form stereo MT movement/stereo Livingston and Hubel Neural Architecture form form color color movement movement 1 2&3 4A Visual Area 2 4B 4C 5 6 Visual Area 1 (Primary Visual Cortex) Orientation and Size (Gabor primitives) Image segmentation based on texture Vertical fat Gabor filters respond most strongly Horizontal skinny Gabor filters respond most strongly Resulting Segmentation Vector fields like using gabors Pre-Attentive Processing 89 73 905 70 927 94 057 96 297 65 098 29 4 08 02 808 50 808 30 802 80 985 0- 802 80 8 56 78 472 98 872 ty 458 20 209 47 577 20 0 21 78 984 38 90r 45 579 04 560 99 272 18 8 89 75 947 97 902 85 589 25 945 73 979 20 9 Color is Pre-Attentive (Pops out) 89 7 3 905 70 927 94 057 96 297 65 098 29 4 08 02 808 50 8083 0 802 80 985 0- 802 80 8 56 78 472 98 872 ty 458 20 209 47 577 20 0 21 78 984 3 8 90r 45 579 04 560 99 272 18 8 89 75 947 97 902 85 589 25 945 73 979 20 9 Generic Pre-Attentive Experiment Number of irrelevant items varies Pre-attentive 10 msec per item or better. Color Orientation Motion Size Simple shading Conjunction (does not pop out) Semantic Depth of Field Compound features (do not pop out) Surrounded colors do not pop out Laws of pre attentive display Must stand out on some simple dimension color, simple shape = orientation, size motion, depth Lessons for highlighting – one of each Lessons: Highlighting how to make information available to attention Using color Using underlining A flying box leads attention Blinking momentarily attracts attention Motion elicits an orienting response More Pre-Attentive Conjunction (does not pop out) Pre-attentive conjunction Conjunctions of motion and shape do pop out. (color also?) McLeod, P., Driver, J. and Crisp, J. (1988) Visual search for a conjunction of movement and form is parallel. Nature 332, 154-155. Driver, J., MacLeod, P. and Dienes, Z. (1992) Motion coherence and conjunction search: Implications for guided search theory. Perception and Psychophysics. 51, 1, 79-85. MEGraph: Experimental system Allows for various topological range highlighting methods MEGraph Goal from 30 to 2000 nodes Pre-Attentive Channels Form (orientation/size) Color Simple motion/blinking Addition/numerosity (up to 3) Spatial, stereo depth, shading, position Pre-Attentive Conjunctions Stereo and color Color and motion Color and position Shape and position In general: spatial location and some aspect of form Pre-Attentive Lessons Rapid visual search (10 msec/item) Easy to attend to Makes symbols distinct Based on simple visual attributes Faces, etc are not pre-attentive Designing symbols Perceptual Channels Color (3) Shape (size, orient) Motion (2?) Texture (2++) Position (x,y) Spatial Channels Like interferes with like Size contrast effect Orient contrast Size contrast effects can cause errors in information display Chris Weigle: orientation channels for info display Mapping data to display variables Data glyphs Position (2) Orientation (1) Size (spatial frequency) Motion (2)++ Blinking? Star glyph Color (3) Method Note we have the problem of heterogeneity – There is no good solution Starplot glyph Spotfire product Integral and Separable dimension (Garner) Can we read display attributes Independently Holistically Speeded classification task. Sort into two piles on one dimension or another Integral red-green yellow-blue black-white red-green x-size y-size Color orientation Motion Color Separable Lessons for Information Display Orthogonality - use a different channel for a different type of information If you need this use separable challenge If you need to highlight by two properties use separable dimensions. The programmable filter We can only look for patterns of simple features Conjunctions of shape, color cannot be programmed for parallel search of field Conjunctions of depth/motion and color/shape can be Integral dimensions tend to be seen holistically cannot be separated Separable dimensions tend to be seen separately Searchlight Model of Attention Useful Visual Field of View Visual Search or Monitoring Strategy Eye Movement Control Searchlight Properties Size varies with data density Size varies with stress level Attention operators work within the searchlight beam Attention is a tunable filter Eye movements 3/sec – A series of saccades