Abstract
This paper presents a new image estimation method for removing the image of eyeglasses from facial images of people wearing eyeglasses. Wearing eyeglasses can change one's appearance. It is necessary to estimate eyeglassless facial images from eyeglassed facial images for evaluating impartially characteristics of both eyeglassed and eyeglassless faces. When we distort an eyeglassed facial image using Wire Frame Model into a facial image showing a different expression, the eyeglass frame is deformed. Therefore we need to extract and remove the eyeglass frame region before the distorting. There are some methods for estimating eyeglassless s facial images by interpolating gray levels in the eyeglass frame region. However these results contain unnatural regions and eyeglass shadows. Our method uses basis vectors, calculated by Principal Component Analysis to obtain eyeglassless facial images, which better capture the characteristics of “eyeglassless faces”. The estimated faces appear natural and we numerically evaluate them using their similarity to eyeglassless facial images.
