Automatic Filtering for the Assignment of Indexes
Download as PPTX, PDF0 likes39 views
The document discusses research on how the visual system assigns "indexes" to objects to track their locations over time. Three key points: 1. Assigning indexes to targets based on their color allows for as good or better tracking performance than flashing the targets, even when the color indication interval is reduced to as little as 50ms. 2. Using color as an indicator is a much stronger "index attractor" than flashing. Performance is largely unaffected even at very short color indication intervals. 3. This suggests the visual system can rapidly filter objects based on a pre-determined color top-down, faster than the automatic indexing process. However, more research is needed to understand how indexing works under these conditions and
Automatic Filtering for the Assignment of Indexes
- 1. Automatic Filtering for the Assignment of Indexes Multitons – Counting – locating and various other things
- 2. ‘Filtering Out’ • “The visual system must be able to do the equivalent of ‘filtering out’ some of the potential false alarm causing signals … All we can do in the filtering is attempt to capitalize on some physically specifiable detectable property that is …used to distinguish those from other stimuli.” (Is vision continuous with cognition? Brain Behavior Sciences 1999, 22, 341-423)
- 3. The First Stage • A two-stage model has been proposed for the tracking task, (McKeever, 1991, Sears, 1991 and Sears and Pylyshyn, 1995). • The first stage can be thought of as the assignment of indexes, which is automatic and driven to objects via primitive features
- 4. The Assignment of Indexes • If the assignment of indexes is based on what Muller & Findlay (1988) call the automatic component of attention, then …
- 5. The Binding Problem • Anne Treisman – Some notes • Features of surface-defining properties – luminance, color, texture, motion, stereoscopic depth • Features of shape-defining boundaries – orientation, curvature, and closure – She reviews the relevant research which seems to be in conflict with the notion that binding requires serial processing. – Current Opinion in Neurobiology, 6, 171-178
- 6. Resolution • “One resolution is that parallel search reflects not preattentive processing but global attention to the scene as a whole with no individuation of separate elements.” • Therefore, a discrepant object is seen as a break in the global structure of the three- dimensional array..
- 7. Is Color-Change Primitive? • Jacquie Burkell & Zenon Pylyshyn – “It seems that when (a color change of any kind) is in fact detected it can be perfectly localized, even when the changed item is in now different from its distracters.”
- 8. Location Encoding • “If the indicated location did not match the target location, but fell within one potential display position of the actual target location (i.e., .9 degrees of visual angle), the response was recorded as ‘near’… if not far.” (from Burkell and Pylyshyn)
- 9. Serial and parallel processing of visual feature conjunctions • “If one of the dimensions in a conjunctive search is stereoscopic disparity, a second dimension of either color or motion can be searched in parallel … these conjunctive tasks (were such that) the observer had the distinct impression that each plane could be search effortlessly, in turn.” • Ken Nakayama & Gerald Silverman (Nature, V. 320 20 March 1986 264-265)
- 10. Color is Powerful 1 • Subjects could determine the orientation of the target – i.e., a T amongst Ls when they were cued symbolically by color, as when cued by color at location, and a luminance location cue. – Efficiency of selective attention: selection by colour and location compared Jari Laarni Mika Koski and Gote Nyman (Perception, 1996, V 25, 1401-1418)
- 11. Color is Powerful 2 • What is important here is that subjects could selectively attend to a group of objects that were one color, and perform better on the discrimination and the orientation tasks, when selectively attending to a subset of items that were set apart via color. – Can one pay attention to a particular color Peter Brawn and Robert Snowden (Perception and Psychophysics V 61(5) July 1999, 860-873
- 12. Open Questions • Can we extend the previous research that involves the detection of single items to the detection of multiple items • Does location encoding requires serial visitation? • What is the duration of that initial assignment?
- 13. Some More Questions • What is the capacity of the assignment of indexes? – What does it mean for an index to be assigned? • Does that mean that it can be enumerated? • Does it mean that it can be located? • Conjunctions – What is the influence of a conjunction of features on the assignment of indexes? – What about the search for multiple items in a conjunction search?
- 14. Our results show that when the targets in an MOT task are indicated by temporally making them a different color than the non-targets, observers track just as well (or better) than when they are indicated by flashing the targets.
- 15. This increase in performance holds even when the indication interval is greatly reduced. In fact, in this case, color signaling is noticeably better than flashing. In fact, for the color signaling condition, performance remains relatively unaffected when the indication interval is reduced from 900ms to 50ms!
- 16. This an outstanding, unexpected result. The results clearly show that color signaling is a much stronger INDEX ATTRACTOR than flashing the targets.
- 17. Points to consider: The onset of the color on the targets cannot be useful to the observer because all objects are colored. The only useful information is the color of the targets versus the non-targets. So how does the system assign indexes to the targets under these conditions? Two possibilities: 1. Particular colors automatically attract indexes. By coincidence, we selected the right color. 2. The visual system is able to perform a fast top-down color filtering for specific pre-determined colors.
- 18. Point 1: Is not supported by data. Observers track both target (red or green) and complement set (red or green) equally well. Point 2: How fast is this filtering process? It must be faster than automatic indexing!!! At 50ms, tracking green is 87% as compared to the flash case which is 65%. But …