Feature Integration Theory: Master Map Inefficacy and Conjunction Search

A University of Melbourne APA Short Lab Report Assignment

Under the 2nd-Year Psychology subject "Cognitive Psychology"

Passed with High Distinction (H1)

============ ============ ============

By Benjamin L., written during Semester 1, 2012

Abstract

In Feature Integration Theory, a ‘master map’ facilitates visual search by highlighting objects that contain the target’s features and directs subsequent conjunction search towards these few candidates. This suggests that the master map would be ineffective when targets and distractors share identical features. Under these conditions, this study hypothesised that the master map would be ineffective and conjunction search accuracy would decrease as set size increased. 35 participants performed self-terminating visual search tasks for conjunction targets amidst varying numbers of distractors. Results showed that search accuracy decreased with greater set size. This suggests that the master map is ineffective when target and distractors share identical features.

Tutor's comments (J. Chan)[not Jackie Chan]:

Overall, excellent work Ben - this is a well formulated lab report! Well done on integrating recent research seamlessly! [Note: For this assignment, we had to focus on Feature Integration Theory rather than more recent accounts]

Feature Integration Theory: Master Map Inefficacy and Conjunction Search

            Visual search involves perceiving both an object’s constituent features and their combination (Prinzmetal, 1995). Single features are rarely sufficient for finding objects (Dehaene, 1989) but simultaneously processing all the visual stimuli in everyday life is impossible (Wolfe & Horowitz, 2004). Instead, visual attention mechanisms process most stimuli at a superficial level, while only a few objects undergo extensive analysis. This allows us to perform the many everyday tasks that require visual search (Wolfe, 1993).

            One influential account is Feature Integration Theory (FIT) (Wolfe & Horowitz, 2004). In FIT, stimuli are coded based on distinct dimensions like colour, orientation, and size (Treisman & Gelade, 1980). Dimensions are categories of features that objects may possess, while features are variations within a dimension. For example, the dimension ‘colour’ contains the feature ‘redness’. Focal attention is essential for conjoining these features into a unitary object (Groeger, Clegg, & O’Shea, 2005) and individuals with impaired visual attention are unable to do this (Arguin, Joanette, & Cavanagh, 1993).

            Feature searches locate targets that are distinguished by a single perceptual dimension (Chan & Hayward, 2009). Features are processed automatically, pre-attentively, and in parallel across vision (Quinlan, 2003), making feature search efficient and independent of the number of stimuli present (set size) (Treisman & Gelade, 1980). Conjunction searches locate targets that are distinguished by conjunctions of multiple features (Wolfe, 1993). As focal attention is required to “glue” features into conjunctions, conjunction search must serially attend to each possible target in turn (Treisman & Gelade, 1980). This serial process makes it is less efficient and search times increase with set size (Chan & Hayward, 2009).

            FIT was challenged by reports of efficient conjunction search and was subsequently revised to include a “master map” (Chan & Hayward, 2009). A feature search would locate objects that contained the target’s features and plot their location onto a “master map”. Next, the master map directs conjunction search solely towards these few locations, thus resulting in more efficient search times.

            However, the master map would not work if all the objects contain the target’s features. Since the master map is essentially a feature-based mechanism, visual displays where both targets and distractors contain identical features should logically prevent the master map from highlighting possible targets. Subsequent conjunction search would have to scan objects without prior guidance until the target is found. Given that conjunction search proceeds serially (Treisman & Gelade, 1980), the time taken to locate a target should increase with set size and if the visual scene is only briefly present, individuals may not have enough time to accurately locate the target. It is therefore hypothesised that, when targets and distractors have identical features, the master map is ineffective and conjunction search accuracy will decrease as set size increases.

Methods

The stimuli were presented on a projector screen using an overhead projector.  35 observers (24 female, age range 19-25 years old) viewed the stimuli while seated comfortably in a classroom. Each stimulus was presented using Microsoftâ PowerPointâ. The stimulus was presented for approximately 1 second. The background was white and each stimulus comprised four white squares with black borders arranged in a 2x2 grid. Within each square there were either two digits (Condition 1) or twenty-six digits (Condition 2). The digits were always black. The target (the number “2”) was always present in one (and only one) square. All the other digits were always the number “5” (the distractors). Although the angular size of the digits was not controlled (the observers viewed the stimuli from different distances), the digits were always large enough to be easily readable.

The numbers “2” and “5” were constructed so that they contained exactly the same line segments (two horizontal line segments and three vertical line segments), albeit arranged differently. Please see Figures 1 and 2.

The observer’s task was to identify which square contained the number 2 (the target).  Each square was uniquely labeled with a letter from the following set {A, B, C, D}. At the end of each trial, the observer wrote down which square they thought contained the target using this alphabetical code.

In each condition, 10 trials were run, alternating between the two conditions. Using an answer sheet, the observer’s average accuracy in each condition was determined. This data was then collapsed across observers and a paired t-test was performed on the resultant data using the software package SPSSâ.

 Figure 1: An example of the stimulus used in Condition 1

Figure 2: An example of the stimulus used in Condition 2

Results

            In this study, the independent variable was set size and the dependent variable was visual search accuracy in terms of the percentage of correct responses. Results are represented in figure 1 below.

Figure 1. Mean Accuracy in Condition 1 & 2 with 95% confidence intervals

            In figure 1, participants in condition 1 had a mean accuracy of 88.85 (SD = 5.29) while those in condition 2 had a mean accuracy of 53.42 (SD = 16.79). A repeated measures t-test revealed that these means were significantly different (t(34) = 13.281, p < 0.001). These results showed that conjunction search in condition 2, which had a larger set size, was significantly less accurate compared to condition 1, which had a smaller set size.

Discussion

            The results above support the hypothesis that, when target and distractor features are identical, the master map is ineffective and conjunction search accuracy decreases as set size increases. This is consistent with previous research which found that conjunction search times increased with set size (Treisman & Gelade, 1980; Dehaene, 1989), thus briefly presented visual displays with large set sizes (i.e. condition 2) may not give participants enough time to correctly locate the target and result in less accurate responses. Conversely, a small set size presented for the same, brief duration (i.e. condition 1) would result in more accurate responses.

            This coheres with descriptions of the master map and conjunction search in FIT. Since the target and distractors in this study consist of identical features – comprising of three horizontal lines and two vertical lines – the ‘2’ target is only distinguishable through a conjunction search that identifies how its features are conjoined (Wolfe, 1993). However, conjunction search is a serial process and attention must be focused on each object in turn until the target is found (Treisman & Gelade, 1980). If the display is only presented briefly, participants may not have enough time to correctly locate the target and produce inaccurate responses.

            When some - but not all the distractors - share the target’s features, as in this study, the master map cannot facilitate conjunction search. The map functions by using an automatic, preattentive feature search to highlight candidate objects containing the target’s features and and subsequent conjunction search focuses on these few candidates, thus reducing overall search times (Chan & Hayward, 2009). This pre-selection of candidates cannot occur when all the distractors contain the target’s features, and hence the master map will not be able to reduce search times.

            However, this study’s findings are not immediately transferable to naturalistic settings. The simple character of the stimuli used in this study are useful for controlled experiments, but everyday visual search tasks often involve more complex scenes than those in laboratory tests (Wolfe, 1993). Future FIT research could test if the results of this study are valid with more naturalistic stimuli.

            In conclusion, this study finds that, when targets and distractors have identical features, the master map is ineffective and conjunction search accuracy will decrease as set size increases. Without the master map, conjunction search remains time-consuming and in a time-scarce situation, this results in inaccurate responses. Due to the simple character of the stimuli used here, care must be taken in extending these results beyond theoretical contexts. To advance our practical understanding of FIT, future research could examine if these findings still hold when naturalistic stimuli are used instead.

References

Arguin, M., Joanette, Y. & Cavanagh, P. (1993). Visual search for feature and conjunction targets with an attention deficit. Journal of Cognitive Neuroscience, 5 (4), 436-452.

Chan, L.K.H. & Hayward, W.G. (2009). Feature integration theory revisited: Dissociating feature detection and attentional guidance in visual search. Journal of Experimental Psychology, 35 (1), 119-132.

Dehaene, S. (1989). Discriminability and dimensionality effects in visual search for featural conjunctions: A functional pop-out. Perception & Psychophysics, 46 (1), 72-80.

Groeger, J.A., Clegg, B.A. & O’Shea, G. (2005). Conjunction search in simulated railway signals: A cautionary note. Applied Cognitive Psychology, 19, 973-984.

Prinzmetal, W. (1995). Visual feature integration in a world of objects. Current Directions in Psychological Science, 4 (3), 90-94.

Quinlan, P.T. (2003). Visual feature integration theory: Past, present, and future. Psychological Bulletin, 129 (5), 643-673.

Treisman, A.M. (1980). A feature-integration theory of attention. Cognitive Psychology, 12, 97-136.

Wolfe, J.M. (1994). Visual search in continuous, naturalistic stimuli. Vision Research, 34 (9), 1187-1195.

Wolfe, J.M. & Horowitz, T.S. (2004). What attributes guide the deployment of visual attention and how do they do it? Neuroscience, 5, 1-7.