Short-and long-time ageing effects in face recognition

The flow of time, either in the short or in the long, brings significant variations in human faces which often make it more difficult to perform face recognition. In formative years, the change in shape and size in the structure of the face is considerable, whereas after the age of 18 years, mainly the texture of the face changes. In addition to this, we can distinguish between two kinds of aging effects: short time or transient changes, and long-time or permanent changes. The former is experienced when comparing the face appearance of the same subject at close time instants and it can be due to a number of changes in the body as well as in the environment. The latter is the change in the facial structure due to the aging or growth/shrink of the tissues. As a consequence, it becomes a tough task to identify a subject by comparing face images sampled at different times. This chapter analyzes the effects of both short-time and long-time aging on face images and tries to localize the variations in appearance and to quantify the errors in recognition. The developed framework for short-time or transient aging analysis, exploits the concept of distinctiveness of facial features and its temporal evolution. The analysis is performed both at a global and local level to define which facial features are more stable over time. Several experiments are performed on publicly available databases with image sequences densely sampled over a time span of several years. In order to evaluate computational models to either detect or compensate for long-time aging effects, a novel face dataset has been purposively collected, composed of 102 individuals with over 2600 images with age separation and other variations including illumination, pose, and eyeglasses. The purpose of this newly acquired dataset, named Delhi face aging database, is to objectively determine the variations in recognition accuracy due to aging, in real life conditions. In order to facilitate a comparative analysis of different algorithms, all images are annotated. Different algorithms are described and their accuracies are compared.