In [16]

In 16, Takimota et. al. used both the Sobel lter and the Gabor jet in order to distinguish a deep wrinkle from a ne wrinkle. The skin features were extracted with a dierence image between the original image and an averaged image. They also used skin color features in the HSV color space. Feng Gao and Haizhou Ai 4 collected thousands of frontal or near frontal face images and labeled them with a subjective age. The training samples were divided into the following four age groups: 0-1, 2-16, 17-50 and 50+. To get a descriptor vector they used Gabor features. Then they utilized the linear discriminant analysis (LDA) technique 11, to build the classier. They further improved it by implementing a Fuzzy LDA version, arriving at the conclusion that Gabor features combined with LDA is the best choice. Also, Gabor lters are employed in the calculation of Bio- Inspired Features (BIF), which is consistently used for age estimation in recent years 9, 14. BIF feature processes an image using a multi-layer feed-forward model where the first layer convolves the image with a set of Gabor filters from multiple orientations and scales, and the resulting vector is downsized with a pooling step, usually with STD or MAX operators. A simplified version of this model 80 is used in 7, where the authors choose the number of bands and orientations manually.