Congruency effects in face recognition for masked perpetrators with real-life images

Recognising masked faces is a challenge in security contexts such as crowd surveillance and criminal investigations. Congruency-based approaches that emphasise the alignment between the encoded and retrieved facial features can improve masked face recognition. In two experiments (total N = 275), we tested how suboptimal viewing conditions, such as those encountered in real-world scenarios, can affect such congruency effects. Participants completed a face recognition task using low-resolution, naturalistic images. They viewed full or masked faces at encoding and test, resulting in four conditions (full-full, masked-masked, full-masked, masked-full). Recognition accuracy was modest (< 60%), illustrating the challenge of masked face recognition under suboptimal viewing conditions. When encoding the full face of a target, contextual congruency neither improved nor impaired recognition performance. However, when a masked target was encoded, recognition improved with congruency between the encoding and recognition phases in Experiment 1 (but not Experiment 2). Confidence and decision time did not predict accuracy as expected under suboptimal conditions, but this dissociation was reduced under lower memory load in Experiment 2. These findings contribute to a deeper understanding of cognitive flexibility and perceptual adaptation in recognising faces and offer insights for improving law enforcement efficiency and enhancing social security.