Robust representations for face recognition: The power of averages

Abstract

We are able to recognise familiar faces easily across large variations in image quality, though our ability to match unfamiliar faces is strikingly poor. Here we ask how the representation of a face changes as we become familiar with it. We use a simple imageaveraging technique to derive abstract representations of known faces. Using Principal Components Analysis, we show that computational systems based on these averages consistently outperform systems based on collections of instances. Furthermore, the quality of the average improves as more images are used to derive it. These simulations are carried out with famous faces, over which we had no control of superficial image characteristics. We then present data from three experiments demonstrating that image averaging can also improve recognition by human observers. Finally, we describe how PCA on image averages appears to preserve identity-specific face information, while eliminating non-diagnostic pictorial information. We therefore suggest that this is a good candidate for a robust face representation.2 Robust representations for face recognition: the power of averages

Temporal tissue patterns in bone healing of sheep

“…Secondly, the tissue-type maps had to be grouped into maps of similar topology, since averaging over images with different topology is not reasonable. Similar to the procedure used in averaging human facial features, 21 the tissue maps were first classified into different developmental stages according to the observed topology of the different tissues (for the definition of the stages, see the next section).…”

“…In order to summarize the information of a group of animals, typically only the amount of the various tissue types (area fractions) are reported as quantitative outcome. 7,12,[19][20][21] This information is limited for two reasons. Firstly, area fractions are missing important topological information, that is, they do not provide location and distribution information.…”

Can Gaze-Contingent Mirror-Feedback from Unfamiliar Faces alter Self-Recognition?

“…It also provides an account of face learning (see e.g., Burton, Kramer, Ritchie, & Jenkins, in press;Kramer, Ritchie, & Burton, 2015;Leib et al, 2014). Accordingly, the created internal representation of a face is tied in an additive manner to the experience of that identity, whereby every new exposure strengthens its average and leads to a stronger internal representation (Burton et al, 2005(Burton et al, , 2011Jenkins & Burton, 2008). Interestingly, this theoretical approach can also explain two interrelated aspects of self-recognition, namely how a visual representation of the own face is created and how this representation accommodates changes in physical appearance during the lifespan.…”

“…Current theorising suggests one way to operationalize this process could be the creation of face averages, in which different instances of the same face are integrated into a single representation (Burton, Jenkins, Hancock, & White, 2005). In this process, 4 information that is relevant to the identity of a person, and therefore present consistently across encounters, is combined to form a robust facial representation for recognition.…”

Evolving the memory of a criminal’s face: methods to search a face space more effectively

“…[5]) -or different types of descriptive information [6]. Other approaches may combine different viewpoints or different images of the same person [7,8]. Here, we evaluate an enhancement to a face production system that allows witnesses to construct faces of criminals.…”