Abstract

Face perception and identification is a task done routinely by all of us. Considerable research has been carried out psychophysically on the detection and identification of human face and sophisticated models have been developed [2]. A critical issue in processing human faces in a machine vision system is the representation of a face since this determines the amount of storage of facial image. In general, a common technique is storing a 2-dimensional array of intensity values as an image file and then compressing the image (e.g. GIF, JPEG compression standards). It is also possible to degrade the image to lower spatial and grayscale resolution. However, at some point reducing resolution (number of pixel blocks) will be counterproductive due to significant loss of information. Here, we report the results of systematic and extensive psychophysical experiments to determine the precise limits of resolution (both spatial and grayscale) for face perception.

© 1995 Optical Society of America