Linda H. Lidborg¹, Anna Yue Gao¹, A. Mike Burton^2,3, Holger Wiese¹; ¹Durham University, ²University of York, ³Bond University

Most people can recognise thousands of familiar faces, with mix-ups occurring only rarely. This is remarkable, considering that the same face can look very different under varying conditions. For example, we recognise familiar faces despite changes in lighting, viewing angle, facial expression etc. But how much facial identity information is necessary for reliable recognition? This can be tested by using morphing techniques and varying the contribution of each face to the morph. Behavioural findings using such morph images suggest that recognition is possible with a minimal amount – possibly as little as 10% – of a face’s identity information preserved. However, many studies have used either unfamiliar-familiar morphs or morphs based on multiple faces, which reduces identity-specific noise and may facilitate recognition with minimal identity information. Here, we used event-related potentials (ERP) and immediate repetition priming to investigate the amount of identity information required for recognition from morphs across two identities (ID1, ID2). Primes were morphs created from pairs of familiar celebrity faces. For each of 24 celebrity pairs, six morph continua were generated, each with six identity steps in 20% increments (0% ID1/100% ID2 to 100% ID1/0% ID2). Target images immediately followed the primes and displayed full identity information. We observed clear ERP priming effects from 300 ms onwards at occipito-temporal electrodes, with more negative amplitudes for 100%, 80%, and 60% morphs, but no priming effects for 40% and 20% relative to 0% morphs. This suggests that the critical threshold for recognition lies between 60-40% identity information. Moreover, the ‘majority’ identity inhibited the processing of the ‘minority’ identity, consistent with an inhibitory mechanism allowing only the processing of the most salient facial identity.

Acknowledgements: This work was supported by the Economic and Social Research Council (ESRC; grant number ES/X002063/1) to HW and AMB.