Paper Title
Feeding Hand-Crafted Features for Enhancing The Performance of Convolutional Neural Networks Based Age/Gender Estimation

Abstract
The convolutional neural network (CNN) is believed to find the right features for a given problem, and thus the hand-crafted features for the problem have been neglected recently. In this paper, we show that studying and using an appropriate feature for the problem may still be important as they can enhance the performance of CNN-based algorithms. Specifically, we propose age and gender estimation methods based on the CNN, which takes as input the face image and its Gabor filter responses. This is based on the domain knowledge that the Gabor response has been one of the most effective features for the face related problems. Precisely, we first derive several Gabor filters with different parameters and then apply them to the given image to obtain several Gabor responses. The stack of input and its Gabor responses are fed through the $1\times 1$ convolution to the CNN so that the input of CNN is a fusion of the input image and Gabor responses. Experiments show that our method helps CNN to find the face related features at the earlier layers and thus improves its performance. Keywords - Convolutional Neural Network (CNN), Hand-Crafted Features, Age Classification Network.