Towards Real-Time Smile Detection Based on Faster Region Convolutional Neural Network

Abstract

Real-time smile detection from digital images taken in unconstrained conditions of lighting and background is still a challenge and can be applied in many real world applications, such as automatic capturing image on a mobile phone camera whenever smile is detected. Previous works usually consider this problem in two steps separately: face detection and smile detection. In this paper, we propose a new method to speed up computational performance of smile detection algorithm using a specialized architecture of Faster Region Convolutional Neural Network (Faster R-CNN). The evaluation from GENKI- 4K dataset shows that our network gains up to 50% faster inference performance and 2 times faster in training than the original Faster R-CNN with the accuracy of 84.5%, which is acceptable for predicting and classifying smile from given images.