Male/female identification from 8 × 6 very low resolution face images by neural network

摘要

It has been widely believed that a human requires at least 16 × 16 resolution to recognize human faces. This paper reports that a neural network can identify sex with more than 90% accuracy from low frequency components of mosaic 8 × 8 images of the cental part of the human face, which we cannot recognize any more as human faces. The recognition rate agrees with that of traditional skull measurement. The network seems to identify sex based on the face shape, especially orbit and cheek bone shape, possibly via shading.