Visual attention and the perception of features and objects

Canadian Psychology, Jan 1994 by Treisman, Anne

Feature integration theory (Treisman & Gelade, 1980) suggested that simple features are codel in parallel in a number of specialized feature maps, but that focussed attention is required to ensure the correct integration of features to specify objects. This is done by a serial scan through a master - map of locations, giving access to the features currently occupying the attended location. These are then integrated to form a representation of the object in the selected location. The present paper reviewed new evidence that has accrued since 1980 and elaborated the model to accommodate the recent findings. The proposal is that four different mechanisms may be involved in different conditions: selection of locations, directed by an externally controlled window of attention; selection by features through preattentive grouping; selection of objects; and selection for access to responses.

1. The nature of perceptual features has been clarified by experiments suggesting (a) "coarse coding" within dimensions, with just a few values specified for colour, for orientation and for other dimensions, (b) a hierarchical ordering of surface - defining media, each supporting a similar coding of features of shape; (c) separate representations for figures defined by darker and by lighter contrast, with focussed attention required to combine across representations; (d) the idea that feature coding remains parallel and global up to the level that defines surfaces in a three (or perhaps 2 1\2) dimensional world; (e) and finally a distinction between preattention (inaccessible to awareness and to the control of behaviour), inattention (that reflects whatever results of preattentive processing can still be retrieved once attention is redirected), and divided attention (that integrates the preattentive, feature - defined boundaries and allows conscious access to global properties of the display).

2. The coding of locations and conjunctions of features has also been further explored. In order to explain fast or parallel search for conjunction targets without any direct coding of conjunctions as such, I and others (e.g., Nakayama & Silverman, 1986; Wolfe et al., 1989; Treisman, 1990) have proposed a second form of attentional selection, controlled not by an externally directed window of attention but by inhibition from one or more separable feature maps. This allows selection of a subset of elements that are spatially intermingled with others rather than grouped in a single spatial window. It also mediates figure - ground segregation on the basis of simple feature differences. The new model differs from my earlier ones in showing parallel access to the shared location map and to the feature maps, making it consistent with the notion of separate parallel pathways coding "what" and "where" (Ungerleider & Mishkin, 1982). When we are given location information in advance (e.g., a spatial cue to where the target will appear), we use the attention window to restrict which features we select. When we have advance information about the relevant features, we use inhibitory control from the appropriate feature maps. When we have neither, we choose an appropriate scale for the attention window and scan serially through locations.

3. A third form of attentional control may be needed to explain object - based phenomena, such as our ability to track a subset of elements among a larger group of randomly moving identical elements (Pylyshyn & Storm, 1988) and the finding of object - specific priming in displays of moving objects (Kahneman, Treisman, & Gibbs, 1992). We suggested that information about any particular object in the field is collected within an episodic representation which includes all its current arbitrary properties, as they appear in its present location at the present time. We used the analogy of an object file that specifies the current state of a particular object and updates it if necessary when the object moves or changes. Recent experiments with moving displays confirm that attention is needed to obtain object - specific priming, but once established through attention, the object file may be used to maintain attention on randomly moving elements defined as parts of the same object.

4. The experiments leading to the hypotheses outlined so far all involved a high perceptual load. Thus these three forms of attentional control may all determine the extent of perceptual processing of different subsets of visual stimuli. On the other hand, when the load is low (only one or two objects), I suggest that no perceptual selection is possible and that attention takes the form of selection for response whenever there would otherwise be response conflict. The negative priming paradigm is an example of this form of attention. De Schepper and I have recently shown not only that novel, meaningless forms can be registered in a single unattended presentation, but also that they may remain in memory, with some inhibitory tag attached to them, across at least 200 intervening trials with similar novel objects. In this paradigm, attention determines which established object file should currently control the choice of response, without preventing its detailed perceptual analysis or its extended survival in memory.