Facial Age Estimation: Lessons Learned by Humans

Abstract

Facial age estimation is the process where either the exact age of a subject or the age group of a subject is determined based on facial information. Facial estimation could figure in numerous real life applications such as age-adaptive human computer interaction, age-based access control to physical sites and/or internet sites and agebased surveillance systems. Recently the topic of automatic age estimation received increased attention and as a result numerous age estimation algorithms where reported in the literature. Although in some cases the results reported are impressive, more effort is required in developing accurate age estimation algorithms capable of handling unconstrained face images where the lighting, facial orientation and image quality may vary considerably. With our work we aim perform a thorough investigation in the process of facial age estimation as performed by human observers. By analyzing the ways in which humans perceive the age of a person we aim to devise improved algorithms that perform more accurately the process of age estimation. As part of our preliminary work, 25 face images were presented to 36 volunteers who were asked to provide an estimate of the age of the person in the corresponding images. During the process the areas of the images that attracted the attention of the volunteers were recorded using an eye tracker. Eye-fixation data is analyzed by generating statistical models describing the main trends of attention focus adopted by different users. The results of a preliminary analysis in relation to personal characteristics of the volunteers and in relation to characteristics of the faces in the images …