Camouflage assessment: Machine and human

• Published in: Computers in industry 2018-08, Vol.99, p.173-182
• Main Authors: Volonakis, Timothy N., Matthews, Olivia E., Liggins, Eric, Baddeley, Roland J., Scott-Samuel, Nicholas E., Cuthill, Innes C.
• Format: Article
• Language: English
• Subjects: Camouflage assessment; Observer modelling; Visual search
• ISSN: 0166-3615; 1872-6194
• DOI: 10.1016/j.compind.2018.03.013

•A visual recognition system has been designed as a human observer model.•Direct comparison of performance from computational model and human participants at the same task.•Performance data from both model and human observers correlated highly. A vision model is designed using low-level vision principles so that it can perform as a human observer model for camouflage assessment. In a camouflaged-object assessment task, using military patterns in an outdoor environment, human performance at detection and recognition is compared with the human observer model. This involved field data acquisition and subsequent image calibration, a human experiment, and the design of the vision model. Human and machine performance, at recognition and detection, of military patterns in two environments was found to correlate highly. Our model offers an inexpensive, automated, and objective method for the assessment of camouflage where it is impractical, or too expensive, to use human observers to evaluate the conspicuity of a large number of candidate patterns. Furthermore, the method should generalize to the assessment of visual conspicuity in non-military contexts.