Limits...
Beyond gist: strategic and incremental information accumulation for scene categorization.

Malcolm GL, Nuthmann A, Schyns PG - Psychol Sci (2014)

Bottom Line: However, this focus has led to a paucity of research into how scenes are categorized at specific hierarchical levels (e.g., a scene could be a road or more specifically a highway); consequently, research has disregarded a potential diagnostically driven feedback process.By recording where in a scene participants fixated prior to making a basic- or subordinate-level judgment, we identified the scene information accrued when participants made either categorization.Our results demonstrate that during scene processing, a diagnostically driven bidirectional interplay between top-down and bottom-up information facilitates relevant category processing.

View Article: PubMed Central - PubMed

Affiliation: 1The George Washington University.

ABSTRACT
Research on scene categorization generally concentrates on gist processing, particularly the speed and minimal features with which the "story" of a scene can be extracted. However, this focus has led to a paucity of research into how scenes are categorized at specific hierarchical levels (e.g., a scene could be a road or more specifically a highway); consequently, research has disregarded a potential diagnostically driven feedback process. We presented participants with scenes that were low-pass filtered so only their gist was revealed, while a gaze-contingent window provided the fovea with full-resolution details. By recording where in a scene participants fixated prior to making a basic- or subordinate-level judgment, we identified the scene information accrued when participants made either categorization. We observed a feedback process, dependent on categorization level, that systematically accrues sufficient and detailed diagnostic information from the same scene. Our results demonstrate that during scene processing, a diagnostically driven bidirectional interplay between top-down and bottom-up information facilitates relevant category processing.

Show MeSH

Related in: MedlinePlus

Paradigm and results from the validation experiment. Participants were asked to categorize individual scenes, but unlike in the gaze-contingent experiment, each scene contained objects identified as diagnostic at either the basic level (example shown on left) or subordinate level (example shown in middle). These objects were revealed at full resolution while the rest of the scene was low-pass filtered. The graph shows mean accuracy as a function of condition and the type of diagnostic information presented. The dashed line indicates chance performance. Error bars represent ±1 SE.
© Copyright Policy - open-access
Related In: Results  -  Collection

License 1 - License 2 - License 3
getmorefigures.php?uid=PMC4232276&req=5

fig6-0956797614522816: Paradigm and results from the validation experiment. Participants were asked to categorize individual scenes, but unlike in the gaze-contingent experiment, each scene contained objects identified as diagnostic at either the basic level (example shown on left) or subordinate level (example shown in middle). These objects were revealed at full resolution while the rest of the scene was low-pass filtered. The graph shows mean accuracy as a function of condition and the type of diagnostic information presented. The dashed line indicates chance performance. Error bars represent ±1 SE.

Mentions: To verify that these significant objects were diagnostic, we asked 36 new participants to categorize scenes at either the basic or subordinate level using the same paradigm as in the gaze-contingent experiment, except that scenes were now masked after 150 ms, and eye movements were not recorded. During presentation, each scene was low-pass filtered except for the objects previously identified as diagnostic at either the basic or subordinate level, which were shown without filtering (Fig. 6). The two types of diagnostic information (basic vs. subordinate) for each scene were distributed evenly across participants in both categorization groups, creating a 2 × 2 mixed design (because there were only eight scenes per basic category and we were measuring accuracy, stimuli type was collapsed). We predicted that accuracy should be high when making a categorization with diagnostic information at the same level (i.e., subordinate objects with a subordinate judgment and basic objects with a basic judgment). We further predicted that because comparatively fewer objects distinguish one subordinate category from another than distinguish one basic category from another (Rosch et al., 1976), swapping the diagnostic objects (i.e., presenting basic objects for a subordinate judgment and vice versa) should have a minimal effect on basic-level categorization but a larger effect on subordinate judgments.


Beyond gist: strategic and incremental information accumulation for scene categorization.

Malcolm GL, Nuthmann A, Schyns PG - Psychol Sci (2014)

Paradigm and results from the validation experiment. Participants were asked to categorize individual scenes, but unlike in the gaze-contingent experiment, each scene contained objects identified as diagnostic at either the basic level (example shown on left) or subordinate level (example shown in middle). These objects were revealed at full resolution while the rest of the scene was low-pass filtered. The graph shows mean accuracy as a function of condition and the type of diagnostic information presented. The dashed line indicates chance performance. Error bars represent ±1 SE.
© Copyright Policy - open-access
Related In: Results  -  Collection

License 1 - License 2 - License 3
Show All Figures
getmorefigures.php?uid=PMC4232276&req=5

fig6-0956797614522816: Paradigm and results from the validation experiment. Participants were asked to categorize individual scenes, but unlike in the gaze-contingent experiment, each scene contained objects identified as diagnostic at either the basic level (example shown on left) or subordinate level (example shown in middle). These objects were revealed at full resolution while the rest of the scene was low-pass filtered. The graph shows mean accuracy as a function of condition and the type of diagnostic information presented. The dashed line indicates chance performance. Error bars represent ±1 SE.
Mentions: To verify that these significant objects were diagnostic, we asked 36 new participants to categorize scenes at either the basic or subordinate level using the same paradigm as in the gaze-contingent experiment, except that scenes were now masked after 150 ms, and eye movements were not recorded. During presentation, each scene was low-pass filtered except for the objects previously identified as diagnostic at either the basic or subordinate level, which were shown without filtering (Fig. 6). The two types of diagnostic information (basic vs. subordinate) for each scene were distributed evenly across participants in both categorization groups, creating a 2 × 2 mixed design (because there were only eight scenes per basic category and we were measuring accuracy, stimuli type was collapsed). We predicted that accuracy should be high when making a categorization with diagnostic information at the same level (i.e., subordinate objects with a subordinate judgment and basic objects with a basic judgment). We further predicted that because comparatively fewer objects distinguish one subordinate category from another than distinguish one basic category from another (Rosch et al., 1976), swapping the diagnostic objects (i.e., presenting basic objects for a subordinate judgment and vice versa) should have a minimal effect on basic-level categorization but a larger effect on subordinate judgments.

Bottom Line: However, this focus has led to a paucity of research into how scenes are categorized at specific hierarchical levels (e.g., a scene could be a road or more specifically a highway); consequently, research has disregarded a potential diagnostically driven feedback process.By recording where in a scene participants fixated prior to making a basic- or subordinate-level judgment, we identified the scene information accrued when participants made either categorization.Our results demonstrate that during scene processing, a diagnostically driven bidirectional interplay between top-down and bottom-up information facilitates relevant category processing.

View Article: PubMed Central - PubMed

Affiliation: 1The George Washington University.

ABSTRACT
Research on scene categorization generally concentrates on gist processing, particularly the speed and minimal features with which the "story" of a scene can be extracted. However, this focus has led to a paucity of research into how scenes are categorized at specific hierarchical levels (e.g., a scene could be a road or more specifically a highway); consequently, research has disregarded a potential diagnostically driven feedback process. We presented participants with scenes that were low-pass filtered so only their gist was revealed, while a gaze-contingent window provided the fovea with full-resolution details. By recording where in a scene participants fixated prior to making a basic- or subordinate-level judgment, we identified the scene information accrued when participants made either categorization. We observed a feedback process, dependent on categorization level, that systematically accrues sufficient and detailed diagnostic information from the same scene. Our results demonstrate that during scene processing, a diagnostically driven bidirectional interplay between top-down and bottom-up information facilitates relevant category processing.

Show MeSH
Related in: MedlinePlus