Papers Covered Change Blindness Change Blindness Current - PowerPoint PPT Presentation

Papers Covered Change Blindness Change Blindness Current approaches to change blindness Daniel J. Simons.  Visual Cognition 7, 1/2/3 (2000)  Failure to detect scene changes  Large and small scene changes Perception Internal vs. External Information in Visual Perception Ronald   Peripheral objects A. Rensink. Proc. 2nd Int. Symposium on Smart Graphics,  Low interest objects pp 63-70, 2002  Attentional blink CS533C Presentation Visualizing Data with Motion Daniel E. Huber and  Christopher G. Healey. Proc. IEEE Visualization 2005, pp. by Alex Gukov  Head or eye movement – saccade 527-534.  Image flicker Stevens Dot Patterns for 2D Flow Visualization. Laura G.   Obstruction Tateosian, Brent M. Dennis, and Christopher G. Healey.  Movie cut Proc. Applied Perception in Graphics and Visualization (APGV) 2006  Inattentional blindness  Object fade in / fade out Mental Scene Representation Overwriting First Impression Nothing is stored( just-in-time)  Single visual buffer  Create initial model of the scene  Scene indexed for later access How do we store scene details ?  Continuously updated  No need to update until gist changes  Maintain only high level information ( gist )  Visual buffer  Comparisons limited to semantic information  Evidence  Use vision to re-acquire details  Store the entire image  Widely accepted  Test subjects often describe the initial scene. Actor  Evidence  Limited space substitution experiment.  Most tasks operate on a single object. Attention  Refresh process unclear constantly switched.  Virtual model + external lookup  Store semantic representation  Access scene for details  Details may change  Both models support change blindness Nothing is compared Feature combination Coherence Theory Pre-processing  Store all details  Continuously update visual representation  Extends ‘just-in-time’ model  Process image data  Multiple views of the same scene possible  Both views contribute to details  Balances external and internal scene representations  Edges, directions, shapes  Need a ‘reminder’ to check for contradictions  Evidence  Targets parallelism, low storage  Generate proto-objects  Evidence  Eyewitness adds details after being informed of them.  Fast parallel processing  Subjects recalled change details after being notified of the  Detailed entities change. Basketball experiment.  Link to visual position  No temporal reference  Constantly updating Coherence Theory and Change Upper-level Subsystems Subsystem Interaction Blindness Implications for Interfaces  Setting (pre-attentive) Need to construct coherent objects on demand  Changes in current coherent objects  Object representations limited to current task  Non-volatile scene layout, gist  Use non-volatile layout to direct attention  Detectable without rebuilding  Focused activity  Assists coordination  Attentional blink  Increased LOD at points of attention  Directs attention  Representation is lost and rebuilt  Predict or influence attention target  Coherent objects (attentional)  Gradual change  Flicker  Create a persistent representation when focused on an  Initial representation never existed  Pointers, highlights.. object  Predict required LOD  Link to multiple proto-objects  Expected mental model  Maintain task-specific details  Visual transitions  Small number reduces cognitive load  Avoid sharp transitions due to rebuild costs  Mindsight ( pre-attentive change detection)

Critique Visualizing Data with Motion Previous Work Flicker Experiment  Extremely important phenomenon  Multidimensional data sets more common  Detection  Test detection against background flicker  Will help understand fundamental perception mechanisms  Common visualization cues  2-5% frequency difference from background  Coherency  1 o /s speed difference from the background  Theories lack convincing evidence  Color  In phase / out of phase with the background  20 o direction difference from the background  Experiments do not address a specific goal  Texture  Cycle difference  Peripheral objects need greater separation  Experiment results can be interpreted in favour of a  Position  Cycle length specific theory (Basketball case)  Grouping  Shape  Oscillation pattern – must be in phase  Cues available from motion  Notification  Flicker  Direction  Motion encoding superior to color, shape change  Speed Flicker Experiment - Results Direction Experiment Direction Experiment - Results Speed Experiment  Coherency  Test detection against background motion  Absolute direction  Test detection against background motion  Out of phase trials detection error ~50%  Absolute direction  Does not affect detection  Absolute speed  Exception for short cycles - 120ms  Direction difference  Direction difference  Speed difference  Appeared in phase  15 o minimum for low error rate and detection time  Cycle difference, cycle length (coherent trials)  Further difference has little effect  High detection results for all values Speed Experiment - Results Applications Applications Critique  Absolute speed  Can be used to visualize flow fields  Study Does not affect detection   Original data 2D slices of 3D particle positions over  Grid density may affect results  Speed difference time (x,y,t) 0.42 o /s minimum for low error rate and detection time  Multiple target directions   Animate keyframes Further difference has little effect   Technique  Temporal change increases cognitive load  Color may be hard to track over time  Difficult to focus on details Stevens Model for 2D Flow Visualization Idea Stevens Model Stevens Model  Initial Setup  Predict perceived direction for a neighbourhood of dots  Start with a regular dot pattern Segment weight  Apply global transformation  Enumerate line segments in a small neighbourhood  Superimpose two patterns  Calculate segment directions  Glass  Penalize long segments  Resulting pattern identifies the global transform  Select the most common  Stevens direction  Individual dot pairs create perception of local  Repeat for all neighbourhoods direction  Multiple transforms can be detected

Stevens Model 2D Flow Visualization Algorithm Results  Ideal neighbourhood – empirical results  Stevens model estimates perceived direction  Data 2D slices of 3D particle positions over a period of time  6-7 dots per neighbourhood   How can we use it to visualize flow fields ?  Algorithm  Density 0.0085 dots / pixel  Construct a dot neighbourhoods such that the Start with a regular grid   Neighbourhood radius desired direction matches what is perceived Calculate direction error around a single point   16.19 pixels  Desired direction: keyframe data  Implications for visualization algorithm  Perceived direction: Stevens model Move one of the neighbourhood points to decrease  Multiple zoom levels required  error Repeat for all neighbourhoods  Critique  Model Shouldn’t we penalize segments which are too short ?   Algorithm Encodes time dimension without involving cognitive  processing Unexplained data clustering as a visual artifact   More severe if starting with a random field