Learning Pyramidal Hierarchical Features for Neonatal Face Detection

Abstract

Because the facial features of the neonates are quite different from those of adults, most of face detection algorithms designed for adult samples aren't suitable for neonatal face detection. In this paper, a novel neonatal face detection algorithm is put forward, which utilizes convolutional neural network with pyramidal hierarchical features. Firstly, the convolutional neural network is used to extract the implicit features of the normalized neonatal image. Then, multi-scale feature maps are selected to predict and detect different sizes of neonatal faces. Finally, a classifier is used to classify the face or not, and the regressor generates the location of the neonatal face. The experimental results based on neonate facial image database prove that: (1)compared with the traditional face detection algorithms, the detection performance is effectively improved, (2)the network is better promoted by using data augmentation or fine-tuning, and (3) selecting multi-scale feature maps from different layers will give different detection performance. The proposed algorithm is superior to the state-of-the-art algorithms in the accuracy on the neonatal facial image database, while keeping real-time performance.