The Oddity Detection in Diverse Scenes (ODDS) database: Validated real-world scenes for studying anomaly detection

Area of Science:

• Cognitive psychology
• Human-computer interaction
• Computer vision

Background:

• Applied screening tasks like medical imaging and baggage screening present unique search challenges compared to standard laboratory settings.
• Real-world scenes contain unspecified targets, often anomalies or oddities, within noisy yet spatially regular environments, unlike controlled lab stimuli.
• Existing laboratory search paradigms do not adequately replicate the complexities of expert visual search in applied domains.

Purpose of the Study:

• To create a laboratory analogue for applied visual search tasks by developing a database of scenes with subtle, ill-specified 'oddity' targets.
• To investigate the relationship between perceived target subtlety and search performance in realistic scene contexts.
• To provide a valuable research tool for studying visual search in expert domains, focusing on anomaly detection in noisy displays.

Main Methods:

• A database of scenes was created, featuring subtle 'oddity' targets (deformations) hidden within 16 variants of each unedited scene.
• Experiment 1 involved eight volunteers searching thousands of scene variants, followed by subtlety ratings of detected anomalies.
• Experiment 2 replicated the study with a larger group of naive searchers to validate the findings.

Main Results:

• Subtlety ratings of anomalies reliably predicted search performance (accuracy and response times) in Experiment 1.
• Image statistics were less effective than subtlety ratings in predicting search performance.
• Prior subtlety ratings from Experiment 1 reliably predicted search outcomes for naive searchers in Experiment 2.

Conclusions:

• The developed scene database serves as a closer laboratory analogue to applied screening tasks involving the search for ill-specified oddities.
• Perceived target subtlety is a strong predictor of search performance, outperforming image statistics.
• The stimuli are useful for studying expert visual search, particularly anomaly detection in complex, noisy visual environments by novices.