Making recognisable faces

Abstract

When delivering visual content on multiple devices and services, faces can often become unrecognisable. This paper draws together research from across the cognitive psychology literature to argue that faces should be treated as a special case when rendering content. Where available, we suggest methods by which recognition can be imp roved within the constraints of the device and service. Firstly, we review the psychology literature to discuss recognition performance when manipulating the face's scale, colour palette, orientation and motion. Secondly, we consider how characteristics of the individual faces can aide or hinder recognition and how caricature may be applied, especially within crowds, to improve it. Thirdly, we show how context can make even the most abstract faces recognisable. Fourthly, we highlight the challenges of making a good portrait, beyond the criteria of simply being recognisable. Finally, we begin to describe a framework for automatically rendering faces 'smartly', such that they will be most recognisable given the device and service of which they are a part