An Efficient Keyframe Extraction Method in Video based Face Recognition

ABSTRACT Video based Face Recognition (VFR) is significantly more challenging when compared to Still Image-based Face Recognition (SIFR).The objective of this paper is to extract keyframes in video to optimize face recognition in video. In the proposed method, the keyframes are extraction through two step process from which facial features are extracted and are given to convolution neural network. The proposed method is tested with four publicly available datasets: the Multiple Biometric Grand Challenge (MBGC), the Face and Ocular ChallengeSeries (FOCS), the Honda/UCSD and theUMD Comcast10 datasets. The experimental results substantially proved thatthe proposed method achieves 3.89% increase in recognition rate when compared to other methods.