Key Frames Based Video Face Verification
Dasari Subbarao1, G. Sai Ram2
1Dr. Dasari Subbarao, Professor, Department of Engineering and Communication Engineering, Siddhartha Institute of Engineering and Technology, Ibrahimpatnam, Hyderabad, India.
2G. Sai Ram, Assistant Professor, Department of Engineering and Communication Engineering, Siddhartha Institute of Engineering and Technology, Ibrahimpatnam, Hyderabad, India.
Manuscript received on 11 April 2019 | Revised Manuscript received on 16 May 2019 | Manuscript published on 30 May 2019 | PP: 609-614 | Volume-8 Issue-1, May 2019 | Retrieval Number: F2682037619/19©BEIESP
Open Access | Ethics and Policies | Cite | Mendeley | Indexing and Abstracting
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC-BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: The availability and abundance of video capture devices like mobile phones and surveillance cameras have instigated research in video face recognition, which is highly related to the law enforcement applications. And the proposed approaches are giving high accuracy in error rates, the performance at lower false accept rates requires significant improvement. Now here we are proposed face verification algorithm where Wavelet Transform and Entropy are used for frames selection from the video followed by feature extraction using deep learning where we will combine Deep Boltzmann Machine (DBM) and Stacked Denoising Sparse Auto-Encoder (SDSAE) with learnt representations for Face verification. After completion of all these steps finally we obtained verification details by multilayer neural network system (MNNS). The proposed feature richness-based frame selection shows fair performance compared to the other methods namely Random frames or frame selection based on no visual reference image quality measures. The proposed method in this paper shows good performance in face verification. Face verification accuracy of proposed method is about 97% and 95% with false 1% accept rate on point and Shoot(PaS), YouTube Video(YTVF) face databases respectively.
Index Terms: Face Verification, Face Recognition, Feature Frames Selection, Deep Learning, Auto Encoder, Deep Boltzmann Machine.
Scope of the Article: Deep Learning